PHILOSOPHY WITHOUT PSYCHOLOGY: A CASE OF WISHFUL THINKING?

The topic of this lecture is the relation between Wittgenstein’s thought and Wittgensteinian philosophy on the one hand, psychology and psychologism on the other. It will start by clarifying how the label ‘psychologism’ should be understood in this context, opting for a neutral rather than derogatory conception. Next it discusses the relation between Frege’s anti-psychologism and that of the early Wittgenstein. The main focus will be on Wittgenstein’s denial that assertion and 29 Philosophy without Psychology: a Case of Wishful Thinking? judgement are of logical relevance. The final sections turn to Wittgenstein’s later thought. Can it avoid the intrusion of psychology concerning the following areas: – meaning – philosophical psychology – philosophical method? Giving short shrift to psychological notions like understanding, perception, judgement and belief is impossible even in philosophy of language and epistemology, given their connections to notions like meaning and knowledge. If Wittgensteinian philosophizing were a kind of psychotherapy, it would also be wholly irresponsible to ignore psychological theories. Fortunately it isn’t. Nevertheless it is neither feasible nor desirable to insulate the clarification of philosophical problems and contested concepts against empirical considerations. I shall substantiate this claim by looking at the problem of animal minds and the role that abilities play for mental and epistemic phenomena. If conceptual analysis is to serve as an instrument of critical thinking, it had better be impure. „Analiza i Egzystencja” 47 (2019)

Schopenhauer, Frege, Russell, Kraus, Loos, Weininger, Spengler, Sraffa (1977, p. 19). Among these, the great logicians Frege and Russell are the philosophically most relevant. Of the two, Russell may have exerted a greater influence on Wittgenstein, by dint of their extensive and intensive conversations in Cambridge between 1911 and. But Frege occupied pride of place in Wittgenstein's own estimate. He met Frege several times before the war, and they continued to correspond until 1919. The Preface of the Tractatus contains an acknowledgement with an unmistakable ranking: 'I am indebted to Frege's great works and to the writings of my friend Mr. Bertrand Russell for much of the stimulation of my thought.' Even after having abandoned his early philosophy, Wittgenstein continued to admire Frege. In his lectures of 1939, he referred to him as 'a great thinker ' (1939, p. 144). Shortly before his death,Wittgenstein observed that 'Frege's style of writing is sometimes great' and, on reading "On Concept and Object", he exclaimed, 'How I envy Frege! I wish I could have written like that' (1988, pp. xiii-xiv).
Interestingly, however, Wittgenstein had a comparatively low opinion of Der Gedanke, which is widely regarded as the crowning testament of Frege's anti-psychologism. In letters that have since been lost he criticized the work, prompting Frege to respond: Of course I do not take offence at your frankness. But I would like to know what the deep reasons for idealism (tiefe Gründe des Idealismus) are that I am supposed not to have grasped. If I understood you correctly you yourself do not take epistemological idealism to be true. Thereby you acknowledge, I think, that there simply are no deeper reasons for this idealism. The reasons for it can only be apparent, not logical… (Frege, 1989(Frege, , 3.04.1920 After World War II, Wittgenstein still disliked Der Gedanke. When Max Black and Peter Geach prepared their Translations from the Philosophical Writings of Gottlob Frege, Wittgenstein counseled them 'to translate Die Verneinung but not Der Gedanke: that, he considered, was an inferior work -it attacked idealism on its weak side, whereas a worthwhile criticism of idealism would attack it just where it was strongest. Wittgenstein told me that he made this point to Frege in correspondence: Frege could not understand -for him, idealism was the enemy he had long fought, and of course you attack your enemy on his weak side ' (1988, pp. XIII-XIV;see Künne, 2009, pp. 26-34).
This article will compare and contrast Frege's and Wittgenstein's respective attitudes towards psychologism rather than idealism. As we shall see, although these two doctrines often go hand in hand, they by no means coincide (see below and Glock, 2009, pp. 124-128). Furthermore, I shall move beyond historical and exegetical points concerning Frege's and Wittgenstein's respective attitudes towards psychologism, to focus on a substantive issue: to what extent is it possible or advisable to disregard psychological facts in clarifying the logical, semantic and mental concepts that lie at the heart both of Frege's work and that of Wittgenstein, early and late. More specifically, I shall be concerned with the following questions: 1. How should psychologism be understood? 2. What is the relation between Wittgenstein's anti-psychologism and that of Frege? 3. Can conceptual clarification à la Wittgenstein avoid empirical psychology with respect to the following topics a. judgment b. meaning and understanding c. philosophical method? 4. If not, does that constitute a psychologistic aberration? Or is the idea of philosophy unaffected by psychological facts a case of wishful thinking? I shall start by clarifying what psychologism and anti-psychologism amount to.

What is psychologism?
The attempt to protect philosophy from incursions by empirical psychology goes back to Kant.
It is popular to accuse Kant and his nineteenth-century successors of confusing logic not just with metaphysics and epistemology, but also with psychology (Kneale, Kneale, 1984, p. 355;Carl, 1994, chapters 1-2;cf. Dipert, 1998). There is some justification for this picture. Kant's transcendental idealism treats the necessary preconditions of experience as features to which the objects of experience have to conform because they are imposed on them by our cognitive apparatus in the course of processing the incoming data. This transcendental psychology was one of the main sources of nineteenth-century psychologistic logic (another one being associationist and introspectionist psychology), because it suggests that the mind can underpin apparently necessary propositions in mathematics and metaphysics (see Glock, 2003). At the same time, Kant also inaugurated crucial anti-genetic and anti-psychologistic modes of thought. What makes a belief a priori is not determined by the way it is acquired, whether it is innate or learnt, but rather depends on the way it can be verified. Furthermore, Kant distinguished between the question of how we acquire a certain kind of experience or belief (quaestio facti) and the question of what the logical and epistemological status of that experience or belief is (quaestio iuris). By the same token, he separated transcendental philosophy from 'empirical psychology', notably Locke's 'physiology of the human understanding' (see 1989, A 84-5/B 116-7; A ix; 1783, §21a). As regards logic, he insisted on the purity of formal logic -a term he coined. He strictly separated 'pure general logic' from psychology, metaphysics and anthropology. He also insisted on the topic-neutrality and normativity of logical laws (1879, VIII; see Trendelenburg, 1840, p. 35).
Kant thereby inspired an anti-genetic and anti-psychologistic strand in nineteenth century German-language philosophy (Sluga, 1997;Glock, 1999b;Anderson, 2005). It was in this context that the label 'psychololgism' first came into prominence. It was first used by Johann Eduard Erdmann in characterizing the approach of Friedrich Eduard Beneke. Its use is intimately linked to two developments. First, a naturalistic-cum-empiricist backlash against German Idealism; and secondly, the institutional parting of the ways between the nascent discipline of psychology and academic philosophy seeking to secure its purity and priority. Here the battle lines were drawn between Wundt, founder of empirical psychology, and Lotze, champion of philosophy as an independent foundational discipline. In the course of this struggle, philosophy 'cleansed itself' from psychology, not just as far as the phenomenological tradition was concerned, but also within the budding analytic movement; whereas psychology, conversely, established itself as an independent discipline (see Künne, 2010, pp. 342-69;Glock, 2015a;Kusch, 1995, p. 182;Beaney, 2014, p. 33f.).
Anti-psychologists are united in the view that philosophy, its subdisciplines (notably logic and epistemology) and related subjects (especially mathematics), are autonomous, distinct not just from psychology, but also from other natural sciences such as physiology. Beyond this context, the use of the label is extremely diverse in several respects.
The first parameter distinguishing various brands of psychologism is scope. Psychologism could be partial, a position which treats logic and mathematics as part of or founded in empirical psychology (e.g. Mill). Or it can be a general perspective on philosophy and connected sub-disciplines. A second parameter is evaluation. For the most part, the label is used pejoratively, e.g. Erdmann, the Neo-Kantians and Husserl. According to this employment, it is constitutive of psychologism to mistakenly conceive a non-psychological phenomenon or problem as a subject of psychology. Psychologism is psychology in the wrong place. But there is also an approbatory employment. The self-proclaimed psychologism in the Brentano school purports to resist formalistic and aprioristic tendencies because of paying due homage to sciences of the mind. Similarly, the term 'logicism' was first used not for Frege's and Russell's programme of reducing mathematics to logic, but more generally as a contrast to psychologism, notably by proponents of the latter, such as Wundt and Nelson (see Gabriel, 1980). Finally, the appellation can also be neutral. Psychologistic positions on a given subject make substantial use of the concepts, theories and methods of empirical psychology. Husserl purported to use the term in this way, but this may well strike us as disingenuous (1900, p. 52;see Künne, 2010, p. 346).
Thirdly, and most importantly, varieties of psychologism differ over the way in which they contrast psychology and a contested discipline. The alleged difference could be one between (i) what is empirical (a posteriori) and what is non-empirical (a priori); (ii) what is subjective and particular and what is objective and universal; (iii) what is descriptive and causally explanatory and what is prescriptive or normative; (iv) the causes which bring about a belief and the reasons that justify a belief. There is a connection between (iii) and (iv): reasons are generally held to have a normative dimension. But a current debate attests to the fact that the nature and extent of the connection is complex, both from a historical and from a substantive perspective (see Star, 2018). (iii) is linked to a further distinction (see Künne, 2010, pp. 344-345), namely between 'radical psychologism', which reduces a contested discipline to descriptive psychology, and 'moderate psychologism', which bases a contested discipline on prescriptions derived from empirical psychology, 'laws of thought' or Denkgesetze.
Finally, there are different conceptions of the kind of psychology and mental laws on which a contested discipline, in particular logic, is grounded. One option is the transcendental psychology-cum-logic mentioned above in connection with Kant's transcendental idealism. Another is empirical psychology of the mentalist kind pursued by Wundt and Brentano. The two differ over (i). Transcendental psychologism tries to explicate non-or preempirical preconditions for generating experience and its objects; mentalist psychologism records empirical regularities, for instance by way of introspection. But they converge regarding (ii): both are concerned with the minds of individuals which are not accessible from a third-person perspective. A final option is neuro-physiological psychologism. It agrees with the mentalist brand on (i), yet not on (ii). Psychology is empirical, but its topic is part of the objective causal order-the central nervous system. This proto-Quinean picture is psychologistic while at the same time avoiding idealism. Conversely, Hegel's position is idealist yet without being psychologistic. According to 'objective idealism', reality does not boil down to episodes in the minds of individuals; it is intelligible only because it is the manifestation of a divine spirit or rational principle (see Glock, 2008, p. 127f.).

Frege's logical anti-psychologism
Frege's anti-psychologism occupies an intermediate position between being partial and being general. It concerns not just logic and mathematics, but also epistemology, a point less frequently noted. Next, unlike Kant and Husserl, Frege is not concerned primarily with (i), i.e., to resist the reduction of logic to something empirical. Nor does he hold with respect to (iii) that logical laws are normative, contrary to a prevalent misinterpretation. 1 There the fact that how things are is essential to how we should pursue our goals (see Glock, 2003, pp. 131-136).
On a connected issue, however, Frege's position is at least misleading. In Logik, a first draft of a book summarizing his logical insights, he writes: The causes which merely give rise to acts of judgment do so in accordance with psychological laws; they are just as capable of leading to error as of leading to truth; they are indifferent to the contrast between true and false. Logic excludes them from its subject matter (1983, p. 2; my translation; the one in Frege [1979] does not contain the final sentence).
It is unclear whether the 'they' after the first semi-colon refers to the causes or the psychological laws. In the second case, at least, this passage seems to yoke logic with (laws of) truth and psychology with (laws of) 'holding to be true', i.e., belief. But that correlation fails in both directions. On the one hand, conceptual truths about belief, ignorance, error and doubt are essential to epistemology and philosophy of mind. On the other, factual truths about the contrast between what is held/appears to be true and what is true are central to many parts of cognitive science. This holds, for instance, for psychological theories about perceptual illusions and cognitive biases, for evolutionary theories concerning systematic limitations of rationality and for linguistic theories about 'framing'. Such sciences do not just seek truths, but also have the contrast between true and false as one of their topics. In Frege's defence one might invoke the following passage: Error and superstition have causes just as much as correct cognition. Whether what you take for true is false or true, your so taking it comes about in accordance with psychological laws. A derivation from these laws, and an explanation of a mental process that ends in taking something to be true, can never take the place of proving what is taken to be true. (Frege, 1918, p. 58f.) To be sure, psychology needs to explain the genesis of false beliefs just as much as those of true ones. But from that, it does not follow that the contrast plays no role in these disciplines.
Next, one might try to dispel the examples of empirical disciplines concerned with contrasts such as that between truth and falsehood, knowledge and mere belief. After all, according to a popular vision, cognitive science can and should be purely mechanical. It ought to relinquish any reliance on logical, epistemological and mental notions in favour of a neurophysiological or neurocomputational approach. But this would be cold comfort to Frege's rejection of naturalism and psychologism. Furthermore, our established use of mental and epistemic expressions defines the topic of cognitive science, its explananda, even if the explanans should turn out to be purely mechanical.
While this passage does not vindicate the idea that the truth/falsehood contrast is immaterial to psychology, it indicates another link between Frege's anti-psychologism and the issue of normativity (iii), namely via the idea of justification (iv). Frege accepted a distinction between reasons which justify believing that p and the causes of subject s believing that p. And as both the passage from Logik and the one from Der Gedanke show, this dichotomy plays an, often under-appreciated, role in his anti-psychologism. 2 With any psychologistic account to logic "we lose the distinction between the grounds that justify a conviction and the causes that actually produce it" (Frege, 1979, p. 147;see Pfisterer, 2010).
Nevertheless, Frege's main attack on the 'psychological logicians' concerns (ii); it is directed mainly against the threat of subjectivism. Logic is objective and universal. Both features can only be secured, Frege thinks, by admitting that its subject matter -truth-values, thoughts (beurtheilbare Inhalte, Gedanken) and their structure -are neither private ideas (Vorstellungen) in the minds of individuals nor linguistic expressions but abstract entities signified by such expressions. Frege's system was axiomatic: all the truths of the predicate calculus can be derived as theorems from its 'Basic Laws' according to rules of inference. Frege understood the axioms not as analytic consequences of arbitrary definitions, but as self-evident truths about abstract entities such as numbers, concepts and relations which are certified by a 'logical source of knowledge'. These axioms 'contain', in undeveloped form, all the theorems which can be derived from them according to rules of inference (1879, §13;1893, Apd.;1979, pp. 267-279).
The need to ensure the objectivity of logic is also paramount in Frege's conception of thoughts and of sense. He distinguished between Vorstellungen, private ideas in the minds of individuals, and Gedanken, thoughts or propositions which are abstract entities inhabiting a Platonic 'third realm' beyond space, time and causality. His grounds were a) a thought, what someone thinks, is true or false independently of someone thinking it; b) two people can entertain and hold true the same thought; c) thoughts can be communicated (1892, pp. 29-32; 1918/19). Frege derives (a)-(c) from a particular conception of logic. But they are also implied by the established use of 'what someone thinks (asserts, etc.)', as well as of derived notions such as that of the 'content' of assertions. They are also prerequisite for explaining communication, disagreement and argument. Only the sense of a sentence, the thought it expresses, is relevant to what is asserted by uttering it and for what that assertion logically implies. By contrast, the 'colouring' (Färbung) i.e., connotations associated with the sentence, is irrelevant (1892, p. 31; 1918/19, p. 63).

The early Wittgenstein: Anti-psychologism on stilts
Young Ludwig likened the development of function-theoretic logic to the scientific revolution in the 17th century (1913a, p. 3). He took over -and transformed -important elements of Frege's and Russell's logical systems. Moreover, he followed Russell in identifying philosophy with the logical analysis of propositions (1922, 4.003f.). But his 'philosophy of logic', departed radically from his predecessors. With considerable chutzpah he includes their work under the label 'the old logic', and castigates them for having failed to clarify the nature of logic (1922, 4.1121, 4.126; 1913b, p. 93; 1914, p. 109). In trying to fill this lacuna, Wittgenstein confronted three accounts of logical truths. In addition to psychologism and Platonism à la Frege, there was Russell's position. According to Russell, the propositions of logic are supremely general truths about the most pervasive traits of reality. This view is reminiscent of Aristotle's conception of metaphysics as the most general science (1903, pp. 3-9 & 106; 1913, pp. 97-101; 1914, pp. 189-190).
Wittgenstein eschews all three alternatives through a 'reflective turn' in the spirit of Kant (see Glock, 1997). Kant distinguished between 'formal logic', which abstracts from the objects of knowledge, and 'transcendental logic', which investigates preconditions of thinking about objects. The former consists of analytic a priori truths. But there are also synthetic a priori truths in mathematics, metaphysics and the a priori elements of science. They hold true of experience (are synthetic) without being made true by experience (are a priori) because they express necessary preconditions of the possibility of experience. Wittgenstein picked up this idea from Schopenhauer and Hertz, who explained the a priori elements of science by reference to structural features of the way we represent objects. The Tractatus extends this idea to the analytic truths of formal logic, while rejecting the idea of synthetic a priori truths. Necessary propositions are neither statements about the way people actually think, nor about the most pervasive features of reality, nor about a Platonist hinterworld but reflect the conditions for the possibility of empirical representation. In contrast to Kant, these conditions no longer reside in a mental machinery. Logic investigates the nature and limits of thought, because it is in thought that we represent reality. But it does so by drawing limits to the 'linguistic expression of thought ' (1922, Pref.). For Wittgenstein's predecessors, necessary propositions are true descriptions, either of how people think (psychologistic logic), or of relations between abstract entities (Frege), or of the most pervasive features of the universe (Russell). For Wittgenstein, by contrast, the a priori status of logical propositions is due not to the alleged fact that they describe a peculiar reality, but to the fact that they reflect rules for describing empirical reality. Logic embodies the necessary preconditions of symbolic representation. Logical truths are tautologies. They combine non-logical propositions which represent possible states of affairs in such a way that all the empirical information cancels out.
In the context of developing this account of logic, the early Wittgenstein modified, expanded and radicalized Frege's anti-psychologism. But his emphasis is not on the objective status of logic or of truth. Frege had argued at length against making both dependent on what humans believe and how they reason, especially in the Preface to Grundgesetze. By contrast, Wittgenstein takes alethic realism for granted: whether a proposition is true depends on how things are rather than on how humans think that they are. His aim is rather to ensure the a priori status of logic, in particular to purge logic and logicism from empirical assumptions such as Russell's appeal to the axiom of infinity.
Psychology is no nearer related to philosophy, than is any other natural science. The theory of knowledge [Erkenntnistheorie] is the philosophy of psychology. Does not my study of sign-language correspond to the study of thought processes which philosophers held to be so essential to the philosophy of logic? Only they got entangled for the most part in unessential psychological investigations, and there is an analogous danger for my method. (1922, 4.1121) In his attempt to avert that danger, the early Wittgenstein discarded even more phenomena as merely psychological. The most striking cases are judgment and assertion, to which we now turn.

Judgement: preliminary clarification and assertion
It is canonical to distinguish -the faculty of judging (judgment, Urteilskraft) as opposed e.g., to perception, will, imagination, memory -the act of judging that manifests this capacity, an act of accepting, affirming, etc., -the 'product' or 'object' of such an act. But the 'act/object'-'act/product' distinction is misleading. First, concerning 'act', Leibniz maintained that a judgment is a deliberate and explicit mental act (14.04.1704). And according to Frege, a judgment is the response to a question, which in turn is a demand to judge (1918/19, p. 143). This yields the idea that a judgment is a deliberate response to a question. But unlike typical acts, including acts of a linguistic and mental kind (asserting, calculating in the head), judging is not subject to the will. I cannot judge at will that there is more than one even prime number.
Next, concerning 'object' and 'product'. What is judged is designated by noun-clauses of the form 'that p'. Yet that p is not a product of A's judging (by contrast to the token-sentence being produced by A uttering 'p'). It is preferable, therefore, to distinguish between act and content. But 'contents' are objects at most in an attenuated sense, Frege's Platonism notwithstanding (see Glock, 2015b). Furthermore, the trichotomy should be supplemented by the enduring state of holding or accepting that p, which corresponds to 'dispositional' i.e., long-standing rather than 'occurrent' or momentary beliefs. Although they were not the first to do so, Frege and Russell distinguished what is asserted or judged, the proposition or thought -from the act of asserting or judging. Frege introduced the sign '|-' to express the act of judging or acknowledging something to be true. That something -the truthbearer -is the thought expressed by the assertoric sentence following '|-'. Every well-formed formula in his axiomatic system has the form (1) |-p The sentence 'p' is the name of a truth-value. Adding the horizontal 'content-stroke' (Inhaltsstrich, Wagrechte) yields '-p', which ought to correspond to the mere thought that p.
Before passing on to a comparison with Wittgenstein, a substantive point is in order. Taken literally, holding to be true goes beyond mere judgment. Like belief, a judgment can merely be directed at what is or is presumed to be the case or a fact (cf. 1879, §3).
(2) A believes that p.
Holding something to be true is a conceptually more demanding phenomenon.
It involves taking a stance towards a (potential) claim about what is the case being true.
(3) A believes that it is true that p.
Unlike (2), (3) requires A to grasp the concept of truth. Russell took over '|-', calling it the 'assertion-sign', and used it to add the force of 'it is true that' to the unasserted proposition. He held that true propositions have the quality of being asserted in a non-psychological, logical sense (1903, p. 35;1913, p. 107 & 196;Russell, Whitehead, 1910, p. 8 & 92).
In 1911, Wittgenstein seems to have maintained that the only things which exist are 'asserted ', i.e., true propositions (McGuinness, 2005, pp. 89-92). Assuming the Russellian conception of propositions, this anticipates his famous claim that the world is the totality of facts rather than things (1922, 1.1). But by the time of Notes on Logic, Wittgenstein had stopped using 'asserted' as equivalent to 'true'. Indeed, he had come to deny that assertion is logically relevant (1913b, pp. 95-96; 1922, 4.023, 4.063f., 4.442).
Assertion is merely psychological.
[…] The assertion-sign is logically quite without significance. It only shows, in Frege and Whitehead and Russell, that these authors hold the propositions so indicated to be true. "|-" therefore belongs as little to the proposition as (say) the number of the proposition. (1913b, p. 95, 103) The Tractatus makes the same point by reference to Frege: Frege's 'judgment stroke' "|-" is logically altogether insignificant [bedeutungslos]; in Frege (and Russell) it only shows that these authors hold as true the propositions marked this way. A proposition cannot possibly say of itself that it is true.
(4.442, see 3.332; 1913b, p. 96) This passage is sloppy, since Frege's judgment stroke is only the vertical part of '|-' (1893, §5; see Künne, 2009, p. 56). Yet this does not settle the question of what logical significance, if any, should be accorded to the sign in its entirety.
Wittgenstein is right to deny that a sentential sign that purports to say of itself that it is true amounts to a proposition with a sense -a proposition that makes a claim that can be assessed for its truth or falsehood. What claim is being made by 'This sentence is true'? Any attempt to provide a genuine answer to this question leads into a vicious regress.
(4) |-p cannot be glossed as maintaining of (4) itself that it is true.
Extracting a statement from (5) engenders regress. What proposition is said to be true in (5)? The proposition that this proposition is true. Hence, we are led to (6) The proposition that this proposition is true is true.
(7) The proposition that the proposition that this proposition is true is true is true and so on.
If this is along the right lines, the assertion-sign can only serve to indicate that a proposition is judged to be true by the person using it. In that respect, at least, it indeed signifies something 'psychological' -holding to be true or asserting. Frege himself grants as much in a letter to Jourdain written in 1912: 'Judging (or recognizing as true is certainly an inner mental process ' (1980, pp. 78-79). But then again, why should that imply logical or semantic insignificance?
The assertion-sign can mark a conceptual difference, namely between merely considering or entertaining and asserting a thought that p. Dismissing that difference as 'merely pragmatic' would rightly be anathema to the later Wittgenstein. And the same should go for 'merely psychological'.
On the other hand, the potential utility of an assertion-sign does not by itself legitimize the uses to which Frege and Russell put it. In Frege and Russell '|-' serves to mark out a proposition as a premise or conclusion in a proof. This is a useful role. But it is not an indispensable one. Context generally suffices to distinguish asserted from unasserted propositions. The need for a special device arises only if, like Frege and Russell, one seeks an artificial language that aspires to dispense with contextual cues. Furthermore, it is a moot question, Wittgensteinian in spirit (1953, § 23-24) and explicitly raised by Davidson (2001, pp. 110-115, but cp. Glock, 2003, whether any sign, any force-indicator, can guarantee that the utterance is meant and / or understood as having a particular force like that of assertion. Finally, in ordinary parlance, at any rate, one cannot infer or draw inferences from premises that one regards as false. Yet Wittgenstein is obviously right to point out that 'we can draw inferences from a false proposition ' (4.023; 1914, 20.10.14). Unfortunately, this is something Frege explicitly denied: 'We cannot infer anything from a false thought ' (1918/19, p. 145); and 'before acknowledging its truth, one cannot use a thought as premise of an inference ' (1918/19, p. 145; 1923, p. 47). Various commentators have defended Frege on the grounds that in the passages at issue by 'Schluss' he in effect means proof, a valid inference with nothing but true premises, that thereby guarantees the truth of its conclusion (Anscombe, 1996, p. 115;Künne, 2009, p. 56;Pfisterer, 2009, chapter 2;Textor, 2011, pp. 80-81). But first, Frege does not say so in the published writings accessible to Wittgenstein. Secondly, even in the passages invoked in his defence (1979, pp. 174-175; 1980, pp. 16-17) he does not explain that expressions such as 'inference' or 'derivation' are to mean proof. Thirdly, even as regards proof, Frege's claim stands in need of further defence, since indirect proof proceeds from propositions which are neither true nor regarded as true. At the same time, Frege explicitly addresses the issue of indirect proofs (1983, p. 264; 1919, p. 146; see also 1879, §4). He recasts the assumption P to be refuted as the antecedent of a conditional. With the help of contraposition, he can therefore reformu late any indirect proof.
(11)|-¬ P [(9) and (10), modus ponens] However, fourthly, one should not run together the notion of inference and of proof, which fulfill two distinct and important roles not just in formal logic and mathematics but in other forms of argument as well.
In the final reckoning, the difference between the early Wittgenstein and his predecessors regarding the assertion-sign is part of a larger contrast concerning their respective conceptions of logic (see Glock, 1996, p. 216-220). '|-' is part of the conceptual notations of Frege and Russell, which stand in the service of their axiomatic presentations of logic, in which one deduces theorems from axioms that have to be not just true but self-evident. Wittgenstein's rejection of '|-' is equally part of his contrasting conceptual notation. In that notation, truth-tables serve not to define the logical constants, but as a way of writing down propositions (4.442; see 3.325) in a way which displays their logical relations -without the need for an axiomatic structure.
As a result, for Wittgenstein, logic is exclusively concerned with the unasserted proposition, which depicts how things are if it is true. At the same time, however, he seems to have concurred with Frege and Russell that such a proposition or picture can be common to the assertion that p, the question of whether it is the case that p, the command to make it the case that p, etc. (4.022; see also 1933, p. 149). This does not imply an inconsistency in his position; however, since logic is exclusively concerned with this common element rather than the diverse mental or linguistic acts. "Judgement, question and command are all on the same level. What interests logic in them is only the unasserted proposition" (1913, p. 96).
More worrying is the claim that an elementary proposition 'asserts the obtaining of a state of affairs' (4.21; see also 4.122, see 4.064). What elementary propositions do can hardly be logically insignificant. That apart, however, the Tractatus position is coherent. The sense of a proposition is 'what it represents', namely a possible state of affairs or situation, an arrangement of objects which may or may not obtain, depending on whether it is true or false. The proposition shows its sense, i.e., 'how things stand if it is true. And it says that they do so stand' (4.022, see 2.201ff., 4.0621). It is reasonable to assume that asserting is the same as affirming or saying that. And in that case, one can simply distinguish between what an elementary proposition shows, its sense, and what it does, namely asserting that this sense -a possible state of affairs -obtains, which comes to the same thing as asserting the obtaining of that state of affairs.

Understanding
Frege showed that the sense of a sentence, the thought it expresses, cannot be private. He concluded that it is an abstract entity which can be apprehended by different people. However, he was forced to supplement this Platonist conception of meaning by a mentalist account of understanding. To understand a sentence is to 'grasp its sense', i.e., to latch on to this abstract entity. In communication the speaker does not induce in the hearer a qualitatively identical idea, but brings him to grasp a numerically identical thought. Understanding is a 'mental process', albeit one at the 'very confines of the mental', since it has to cross the ontological gap between the mental and the abstract. The nature of this process remains a mystery. It is equally mysterious how we can check whether speaker and hearer have indeed latched on to the same abstract entity, since Frege accepts the received idea that the contents of the mind are private (1892, pp. 29-30; 1918, p. 68; 1893, §32; 1979, pp. 137-145). The early Wittgenstein combined Frege's anti-psychologistic evasiveness with Russell's logical atomism. We are capable of constructing and understanding an unlimited number of propositions because we can calculate their senses on the basis of knowing their constituents and their mode of combination. This implies that understanding is a process of calculation. The sense of a molecular proposition is derived from that of its constituent elementary propositions according to the rules of truth-functional combination. The sense of elementary propositions is derived from the meaning of its unanalyzable elements -logically proper names. The process of calculation presupposes a process of analysis, since the constituents and logical forms of ordinary propositions are hidden behind their grammatical surfaces (4.002; 4.024-6, 3.318). Both processes must be unconscious: we are usually not aware of them, and they will only be made explicit by a successfully-completed logical analysis of the propositions of natural languages. The result of constructing (speaker) or calculating (hearer) the sense of a proposition is a string of 'thoughts' which accompany communication. Thoughts are psychic facts which consist of thought-constituents that correspond to the names in the propositional sign. The relation of these constituents to the objects of the depicted situation 'would be a matter for psychology to find out'. More generally, the study of 'thought-processes' is irrelevant to logic (4.1121;1974a, letter to Russell 19.08.1919, 10.11.1914.
In sum, the early Wittgenstein's account of meaning features an unfortunate combination of avowed anti-psychologism and closet psychologism-cum-mentalism. Bracketing psychological notions in the name of anti-psychologism is unwarranted and counter-productive, even in the philosophy of language and epistemology, for these notions are internally connected to notions such as meaning and knowledge.
Wittgenstein's later approach is radically different. Instead of sweeping the problem of how we explain and understand words and sentences under the carpet in the name of anti-psychologism, he develops a non-psychologistic account of understanding. He rejects the assumption shared by mentalism and Platonism, namely that sentences merely provide the perceptible clothing of language-independent thoughts, an assumption Frege expressed by writing "The thought … gets clothed in the perceptible garb of a sentence" (1918, p. 61).
Frege and the early Wittgenstein were right to regard mental processes and images as irrelevant to sentence meaning, but wrong to think that the same goes for the concept of understanding.
The meaning of an expression e cannot transcend the understanding of competent speakers. It is immanent rather than "hidden" (1953, §126-128). It cannot be at odds with explanations of e which competent speakers can proffer on reflection or at least accept when they are formulated by experts. '"The meaning of a word is what the explanation of meaning explains." I.e. if you want to understand the use of the word "meaning," look at what is called "explanation of meaning" ' (1953, §560). The meaning of e is also determined by how competent speakers understand and explain e. Understanding is a 'correlate' of explanation and meaning, and instead of asking, 'What is the meaning of e?' we should ask, 'How is e explained?' and, 'What are our criteria for someone understanding e? ' (1979, p. 43;1933, p. 11;1974a, pp. 45 & 60;see Glock, forthcoming).

Concluding thoughts
According to the enlightened anti-psychologism of the later Wittgenstein, communication is not a matter of making something happen in the hearer's mind -the grasping of a sense -such that it is irrelevant what happens thereafter. Understanding an utterance does not consist in having an experience, nor is it anything else which crosses the hearer's mind. Rather, it is a capacity, which is manifest in how the hearer reacts to the utterance (1953, §317, p. 363, pp. 501-510). Understanding a word is also an ability, which manifests itself in three ways: i) how one uses it; ii) how one responds to its use by others; iii) how one explains what it means when asked (1953, §75; 1979, p. 48ff.; 1976, pp. 19-28). These three criteria of understanding can in principle come apart (someone might use a word correctly without reacting appropriately or being able to explain it). But it is crucial to our concept that they commonly coincide. More generally, the mind is a complex of abilities. But in establishing what abilities, and how they relate to each other, we must take into account psychological facts. That at any rate is what I have argued elsewhere (Glock, 2017a). On this occasion I end with a brief observation on another potential relation between psychology and philosophy. More strictly speaking, it concerns the relation between psychology and philosophical method or 'metaphilosophy'. Wittgenstein's later work features extensive reflections on the nature of philosophy and its problems. These revolve around what one might call a 'phenomenology of philosophical puzzlement'. Wittgenstein highlights the peculiar difficulties we encounter in philosophical trains of thought, and how these contrast with challenges in other intellectual endeavors. These observations furnish a novel understanding of the character of philosophical problems, which in turn calls out for an approach that contrasts sharply with scientific theory-building (see 1980, p. 1; 1974b, p. 193; 1979, pp. 27-28; 1954/1955, pp. 113-114).
According to irrationalist interpretations, the approach guiding Wittgenstein's philosophizing is a kind of psychotherapy modeled on psychoanalysis. In that case it would also be wholly irresponsible to ignore industrial strength psychological theories about the cognitive and affective mechanisms that make subjects susceptible to the disease. Fortunately, Wittgensteinian philosophy is not a psychotherapeutic attempt to cure people from the urge to philosophize, for his phenomenology reveals that philosophical puzzles are conceptual. As such they must be resolved by way of conceptual clarification and informal logic ('grammatical investigations'). We need to understand the logical structure of problems and arguments rather than the psychological profile of those propounding them (see Glock, 2017b). What I have argued just now is that this does not exonerate us from clarifying mental notions and paying heed to psychological facts and theories. The inclination to think otherwise would indeed amount to a case of wishful thinking. 3 Frege, G. (1884). Die Grundlagen der Arithmetik. Breslau: Koebner. Giving short shrift to psychological notions like understanding, perception, judgement and belief is impossible even in philosophy of language and epistemology, given their connections to notions like meaning and knowledge. If Wittgensteinian philosophizing were a kind of psychotherapy, it would also be wholly irresponsible to ignore psychological theories. Fortunately it isn't. Nevertheless it is neither feasible nor desirable to insulate the clarification of philosophical problems and contested concepts against empirical considerations. I shall substantiate this claim by looking at the problem of animal minds and the role that abilities play for mental and epistemic phenomena. If conceptual analysis is to serve as an instrument of critical thinking, it had better be impure.

Introduction
What is it that is true or false in a primitive, non-derivate way? According to Frege, early Russell, and much of subsequent analytic philosophy, the primary truth-bearers are mind-independent entities. Mental or linguistic acts, such as judgments and assertions, can be said to be true or false -if at all -only in a derivative sense. In recent years, this entity-based approach has been challenged. Some have proposed act-theoretic accounts identifying #0# * Silver Bronzo -Assistant Professor of Philosophy at the National Research University Higher School of Economics, Moscow. He received his PhD in 2015 from the University of Chicago and has published several articles and book chapters in the history of analytic philosophy and the philosophy of language.
Address for correspondence: National Research University Higher School of Economics, School of Philosophy, Staraya Basmannaya Ulitsa 21/4, Moscow, 105066 Russia. E-mail: silver.bronzo@gmail.com. the primary truth-bearers with concrete mental or linguistic acts. 1 Others have proposed accounts that are not act-theoretic, because they deny that acts are true or false, but are nonetheless act-based, because they hold that the primary truth-bearers essentially involve concrete acts. In particular, some have argued that the primary truth-bearers are the "products" of mental or linguistic acts. 2 I am going to argue that Wittgenstein's Tractatus proposes an act-based account and is in this respect deeply opposed to Frege. For the Tractatus, the primary bearers of truth and falsity are facts-in-use. These essentially involve concrete acts, namely acts of using mental or linguistic facts in a certain way. I shall leave open the question of whether the Tractatus' account is not only act-based, but also act-theoretic. This depends on whether a fact-in-use (i.e., a fact used on a certain occasion in a certain way) is the same as the use of a fact (i.e., the concrete act of using a fact in a certain way) -a question that I shall not attempt to answer. 3 Three platitudes I am going to contrast the Tractarian account of truth-bearers with Frege's by looking at how they deal with three platitudes that are widely believed to place constraints on any adequate view of the matter. 4 First, what I shall call Repeatability. One can say or think the same thing on different occasions. Suppose I say, now, that 2 plus 2 equals 4. This is a particular, historical event. But I can say the same thing again -here I go: 2 plus 2 equals 4. I just performed two distinct historical acts of saying, but I said the same thing. Secondly, what I shall call Shareability. Different persons can say or think the same thing. Suppose I say, again, that 2 plus 2 equals 4. You can say the same thing. My act and your act are different historical events, but they amount to the saying of the same thing. Third, what I shall call Objectivity. What we say or think may be true independently of the fact that anybody has ever said or thought it. If I say that 2 plus 2 equals 4, I say something true; but what I say would have been true even if I had never said or thought it. It was true before humankind even existed, and will continue to be true when humankind becomes extinct and there is no one left in the universe to think or say anything at all.
Let's see how the theory that emerges from a major strand in Frege's mature writings -including Der Gedanke -purports to vindicate these platitudes. For the sake of brevity, I will refer to it simply as "Frege's theory", even though, as we shall see below, there are aspects of Frege's philosophy to pull in a different direction.

Frege's theory
Shortly after the beginning of Der Gedanke, Frege sets out to "delimit […] the region within which truth can be predicated, the region in which there is any question of truth" (1918, p. 59/326). His answer is that "the only thing that raises the question of truth at all" is a Gedanke, a "thought" (p. 60/327). Thoughts are expressed by sentences and constitute their senses (p. 61/328); they are grasped in acts of thinking (p. 62/329); and are acknowledged as true in acts of judgment (p. 62/329). They are abstract, mind-independent entities inhabiting a "third realm" (p. 69/337) distinct both from the "outer" and the "inner world" (p. 66/334). The outer world is the realm of what can be perceived by the senses, i.e., physical objects such as "trees, stones, and houses". The inner world is the realm of "ideas", i.e., mental items that always belong to the content of someone's consciousness and include "sense impressions", "creations of this imagination", "sensations", "feelings", "moods", "inclinations" and "wishes" (p. 67/334).
Thoughts are in some respects like physical objects and unlike ideas; in some respects like ideas and unlike physical objects; and in some respects unlike both. They share with physical objects the fact that they are mind-independent. Thoughts may exist and be true even though nobody thinks them: 5 [T]he thought we have expressed in the Pythagorean theorem is timelessly true, true independently of whether anyone takes it to be true. It needs no owner. It is not true only from the time when it is discovered; just as a planet, even before anyone saw it, was in interaction with other planets. (Frege, 1918, p. 69/337) In order to be true, thoughts -e.g. laws of nature -not only do not need to be recognized by us as true: they do not have to have been thought by us at all. A law of nature is not invented by us, but discovered, just as a desolate island in the Arctic Ocean was there long before anyone set eyes on it. (Frege, 1897, p. 144/233) The mind-independence of thoughts sets them apart from ideas, which always need an owner. An idea is always the idea of someone who has it; they don't "go around the world without an owner, independently" (1918, p. 67/334). However, thoughts share with ideas the fact that they are not sensibly perceptible. Ideas cannot be seen or perceived, they can only be had. For example, one sees a tree, but has the visual impression of a tree (p. 67/334). Similarly, a thought "is in itself imperceptible by the senses" and becomes sensibly perceptible only in so far as it "gets clothed in the perceptible garb of a sentence" (p. 61/328). Finally, thoughts differ both from physical objects and ideas because they do not belong to the temporal and causal order. Both physical and mental objects are subject to change as a result of changes in other physical or mental objects. For example, a light ray may strike the visual nerve and give rise to an idea of a tree, which may interact with other ideas and lead to a decision, which may in turn lead to a bodily movement (pp. 71/338, 76/334). 6 Thoughts, by contrast, are "timeless and unchangeable" (p. 76/343). Thoughts do not come in and out of existence and if they are true, they are so timelessly. Moreover, they are causally efficacious, or "actual" (wirklich), only in an indirect way, in so far as they are grasped. Being causally efficacious is an external or "inessential" property of thoughts; and even if thoughts can be causally efficacious in an indirect way, they are never in turn subject to change as a result of mental or physical events (pp. 76-77/344-345).
A thought, for Frege, (a) is not the product of a mental act of thinking, and (b) is not identical to a mental act of thinking: Now we cannot [a] regard thinking as a process which generates thoughts. It would be just as wrong to [b] identify a thought with an act of thinking, so that a thought is related to thinking as leap is to leaping […].
[I]f thoughts [a] only came into existence as a result of thinking or [b] if they were constituted by thinking, then the same thought could come into existence, cease to exist, and then come into existence again, which is absurd. As I do not create a tree by looking at it or cause a pencil to come into existence by taking hold of it, neither do I generate a thought by thinking. (Frege, 1897, p. 149/237) Thinking a thought should not be understood on the model of building a house, where the house is the product generated by the activity of building. It should also not be understood on the model of taking a leap, where the leap simply consists in the act of leaping. 7 Rather, thinking a thought should be understood on the model of grabbing a pencil: the pencil is there, independently of the act of grabbing, and you grab it. (Even better, it should be understood on the model of grabbing a stone, since the existence of artifacts such as pencils is ultimately due to actions.) As Frege goes on to say, The metaphors that underlie the expressions we use when we speak of grasping a thought, of conceiving, laying hold of, seizing […] put the matter in essentially the right perspective. What is grasped, taken hold of, is already there and all we do is take possession of it. (Frege, 1897, p. 149/237) Thus Frege explicitly rejects both act-based accounts identifying the primary truth-bearers with the products of acts of thinking and act-theoretic accounts identifying the primary truth-bearers with acts of thinking -and the rejection carries over implicitly to any account that makes the primary truth-bearers dependent on mental and linguistic acts.
The view that the primary truth-bearers are mind-independent entities goes hand in hand, in Frege, with a relational analysis of indirect speech. For Frege, "S thinks/judges/says that p" has a relational form. The that-clause refers to a thought (Frege, 1892, p. 37/160). More specifically, it functions as a "proper name". 8 The thought it refers to is, accordingly, an object. 9 Now, if the that-clause is the proper name of an object and the expression schematically represented by "S" is the proper name of a thinker or speaker, i.e., another object, and if the whole sentence is meant to express a thought (which is obviously the case), then the verb of the main clause can only stand, in Frege's system, for a two-place, first-level relation. And in fact, Frege describes grasping a thought in relational terms: "When [a man] grasps or thinks a thought he does not create it but only comes to stand in a certain relation to what already existed" (1918, p. 70/337, n. E; see also p. 76/344).
It is easy to see how Frege's theory accounts for our three platitudes. A person can think or say the same thing on different occasions (Repeatability), and different people can think or say the same thing (Shareability), for the same reason for which a person can grab the same stone on different occasions, and different people can grab the same stone. Moreover, what we say or think may be true or false independently of the fact that anybody has ever thought or said it (Objectivity), for the same reason for which a stone can weight 1 kg independently of the fact that anybody has ever grabbed it. 8 "The subordinate clause [in indirect speech] could be regarded as a noun, indeed one could say: as a proper name [Eigenname] of that thought […] which is represented in the context of the sentence structure" (Frege, 1892, p. 39/162). 9 For Frege, as is well known, a proper name (Eigenname) is any expression, simple or complex, which refers to an object, "this word taken in the widest range", so as not to be restricted to what belongs to the temporal and causal order, let alone to what is material and spatial (Frege, 1892, p. 27/153).

A questionable reading of Frege?
In taking at face value Frege's talk of the "third realm", I disagree with some of the claims of anti-metaphysical readers of Frege. 10 There are in fact different anti-metaphysical readings of Frege and different points of disagreement between each anti-metaphysical reading and its opponents. This is not the place to try to resolve a complex debate that has gone on for more than three decades. I will only add some remarks that help to locate the interpretation I endorse in the extant debate and provide some reasons for preferring it to one of the most relevant alternatives.
A significant part of the disagreement between the anti-metaphysical reading proposed by Joan Weiner (1990Weiner ( , 1995aWeiner ( , 1995b and the Platonist reading proposed by Tyler Burge (1992) concerns the issue of whether thoughts, for Frege, are essentially capable of being grasped and expressed. Weiner agrees with Burge that thoughts, for Frege, are independent of our actual mental or linguistic performances, but maintains, contra Burge, that they are not independent of our possible mental or linguistic performances (1995a, p. 591; 1995b, pp. 368-72). In this sense, she insists, they are minddependent. The interpretation I presented in the previous section leaves open the question of whether Fregean thoughts are mind-dependent in this sense. My contention is that they are independent of our actual mental or linguistic performances. This is the sort of mind-independence that I am interested to discuss in this paper and with respect to which I want to draw a contrast between Frege and the Tractatus.
There is, however, a substantive disagreement between the interpretation I presented and the deflationist brand of anti-metaphysical reading advocated by Thomas Ricketts (1986). On Ricketts' view, the whole talk of thoughts as objects inhabiting a third realm is only a way of "systematically redescribing" our practicing of asserting and inferring (1986, pp. 72, 92). In my terms, such talk is not meant to express a philosophical account of the sort of platitudes with which I started, but a way of restating those platitudes. For example, the claim that thoughts are objects existing independently of the fact that anybody grasps them is just a way of saying that different people can think the same thing, and that something can be true even if nobody thinks it. Frege's claims about thoughts and the third realm are simply meant to restate the data, not to explain them.
What I find problematic in this reading is that it does not specify what is supposed to be achieved by Frege's "redescriptions". Ricketts speaks of "systematic" (1986, pp. 72, 92) or "structured" (p. 73) redescriptions. But nowhere does he explain what the relevant sort of systematicity is supposed to be. On his reading, Frege's redescriptions cannot be "systematic" in the sense that they offer a unified explanation of the relevant platitudes based on a relational analysis of indirect speech and an ontology of thoughts as mind-independent entities. In which sense, then, are they "systematic"? What kind of systematicity do we gain, for instance, in restating "Different people can state the same thing" as "Different people can stand in a relation of 'grasping' to the same atemporal, mind-independent object"? The only systematicity I can discern in such redescriptions, on the assumption that they are not meant to provide any sort of explanation, is that they are systematically misleading, since they suggest precisely the sort of explanation that, according to Ricketts, Frege did not really endorse.
None of this is to deny that there are several aspects of Frege's philosophy -rightly emphasized by Ricketts and other anti-metaphysical readers -that are in tension with the view I ascribed to him in the previous section. These include, arguably, Frege's claim that the distinctive character of his logic lies in the fact that he "start[s] from judgments and their contents, and not from concepts" (Frege, 1880/81, p. ,17/16;cf. Ricketts, 1986, p. 67); his reluctance to call truth a property (Frege, 1918, p. 62/329;cf. Ricketts, 1986, p. 79); the idea that "what logic is really concerned with is not contained in the word 'true' at all but in the assertoric force with which a sentence is uttered" (Frege, 1915, p. 272/323;cf. Ricketts, 1986, p. 84); and the definition of sense in terms of linguistic expressions and their reference (Frege, 1893, §32;cf. Dummett, 1986). What I deny is that the passages I discussed in the previous section can be reconciled with the perspective that emerges from these other regions of Frege's philosophy. 11 Now I am going to argue that the Tractatus rejects the major tenets of the theory I ascribed to Frege in the previous section. For the Tractatus, (i) the primary truth-bearers are not mind-independent, (ii) can be sensibly perceptible, (iii) are not objects, and (iv) indirect speech is not liable to a relational analysis.

The Tractatus on pictures, thoughts, and propositions
The Tractatus ascribes truth and falsity to three sorts of items: "pictures", "thoughts", and "propositions". Here are some representative passages: 12 The Tractatus speaks of truth also in two other contexts: in connection with solipsism ("This […] raises the question, to which extent solipsism is a truth," 5.62), and in relation to the views apparently expressed in the Tractatus itself ("The truth of the thoughts expressed here seems to me unassailable and definitive," Preface). These cases, however, raise special problems and I will ignore them in what follows. Given what the Tractatus says about its own propositions (namely, that they are nonsensical, 6.54) and about philosophy (namely, that "it is not a theory, but an activity", and that "the result of philosophy is not a number of 'philosophical propositions", 4.112), it is reasonably clear that the term "truth" is used in these contexts in a special sense, which differs from the sense it carries elsewhere in the book.
So let's focus on pictures, thoughts, and propositions. How are they related? This is a very controversial exegetical issue. In fact, it is not obvious that there is room for any coherent account of the matter that does not do some violence to (or at least some massaging of) the relevant texts. I will begin with some points that I take to be uncontroversial.
It is clear that thoughts and propositions are pictures: 3 The logical picture of facts is the thought. 4.01 The proposition is a picture of reality.
Thoughts and propositions fall, therefore, within the scope of the socalled "Picture Theory". It is also reasonably clear that the Picture Theory includes the following tenets. A Tractarian picture consists in the fact that a certain number of elements stand to one another in a certain way. Each element of the picture stands for, or "deputizes", an element of reality. The fact that the elements of the picture stand to one another in a certain way represents the fact that the correlative elements of reality stand in that way. If they do indeed so stand, the picture is true; otherwise it is false. Finally, the picture and what it pictures share the same form, which means that their elements have the same combinatorial possibilities (2.1-2.225).
Things get more complicated when we try to be more specific about the relation between pictures, thoughts, and proposition. With regard to the relation between pictures and thoughts, there are two live options. One option is to maintain that pictures and thoughts are co-extensional, as suggested by the following passages: 3 The logical picture of facts is the thought. 2.182 Every picture is also a logical picture.
(On the other hand, for example, not every picture is spatial.) If a thought is a logical picture, and every picture is (inter alia) a logical picture, then -it would seem -every picture is a thought. And of course, since a thought is a logical picture, every thought is a picture. A thought, on this reading, would be a picture considered only as a logical picture, in abstraction from the fact that it might also be a more specific sort of picture, such as a spatial or temporal picture. Thus spatial or temporal pictures, which represents reality in virtue of sharing with it not only logical form but also spatial or temporal form, count as thoughts. 13 The other option is to hold that thoughts are only-logical pictures, namely pictures that share with what they represent nothing more than logical form. On this reading, thoughts are a proper subset of pictures. There are pictures, such as spatial or temporal pictures, that are not thoughts. 14 It seems to me that this second reading makes best overall sense of the text. A piece of textual evidence in its support is the following. The Tractatus says that "the logical picture can depict the world" (2.19), where "the world is everything that is the case" (1). It also says, in accordance with its definition of a thought as a logical picture, that thoughts can depict the world: "The totality of true thoughts is a picture of the world" (3.01). But for the Tractatus, it is not the case that any sort of picture can depict the world. For example, a spatial picture, qua spatial picture, can only depict a spatial situation: it cannot depict, say, a temporal or chromatic situation (2.171). It cannot do so because it does not share the form of a temporal or chromatic situation. The only form that belongs to everything that is the case is logical form (2.18). Thus the only sort of picture that can depict everything that is the case is a picture that represents simply in virtue of its logical form. It is, in other words, an only-logical picture. So "logical pictures" and thus "thoughts" are only-logical pictures.
This conclusion fits well with the fact that logical pictures are ostensibly introduced as a special case of pictures: 2.181 If the form of representation is the logical form, then the picture is called a logical picture.
If every picture were a logical picture, this definition would make no sense: there would be no condition that a picture would have to satisfy in order to be "called a logical picture". Admittedly, our conclusion does not fit well with the immediately following remark, which has already been quoted: 2.182 Every picture is also a logical picture.
If a logical picture were here an only-logical picture, this statement would be self-contradictory. The "also" and the "only" would clash with each other.
A spatial picture, for example, can't be also an only-logical picture, for to say that it is an only-logical picture is precisely to say that it is nothing more than that. My suggestion is that the Tractatus is in fact working, even at such a minimal textual distance, with two notions of "logical picture": an inclusive notion, which applies to all pictures, and a narrow notion, which applies only to only-logical pictures. Thoughts, which can depict the world, are logical pictures in the narrow sense.
We can now move on to the question of the relation between thoughts and propositions, which is even more complicated. One the one hand, it seems that propositions and thoughts are supposed to be the very same thing: 4 The thought is the significant proposition.
The Tractatus speaks here of "significant propositions" instead of propositions tout court; but on many occasions, it clearly uses the two terms interchangeably -for instance, when it says that "a proposition shows its sense" (4.002). So the Tractatus appears to identify thoughts and propositions. But on the other hand, the identity of propositions and thoughts seems flatly incompatible with what Wittgenstein writes in a famous 1919 letter to Russell. Russell had sent a letter to Wittgenstein asking for several clarifications about the Tractatus. He noticed that, for the Tractatus, "The thought is the significant proposition" (4) and "The totality of propositions is the language" (4.001). So he asked, "Does a Gedanke consist of words?" (Wittgenstein, 2008, p. 96). To which Wittgenstein replied: No! But of psychical constituents that have the same sort of relation to reality as words. What those constituents are I don't know. (Wittgenstein, 2008, p. 99) It is beyond doubt that a proposition, for the Tractatus, consists of wordsor more generally, of linguistic items. Not only propositions are said to be part of language, as Russell notices, but a proposition is said to be a "propositional sign", such as a "sound or written sign" in "its projective relation to the word" (3.11, 3.12). Moreover, the elements of completely analyzed propositions are said to be "simple signs" or "names" (3.202). The passage from the letter to Russell appears therefore to imply that thoughts and propositions form disjoint classes. Both propositions and thoughts are pictures, but thoughts consist of psychic elements, whereas propositions consist of linguistic elements. Thoughts are mental pictures, whereas propositions are linguistic pictures. The idea that thoughts and propositions are different kinds of pictures leads to a particular construal of another Tractarian characterization of their relation. For the Tractatus, a proposition is the sensibly perceptible expression of a thought: 3.1 In the proposition the thought is expressed perceptibly through the senses.
If thoughts and propositions are distinct items, then to be the "expression" of a thought is a matter of being an item that is correlated to another item. Behind every proposition there is a thought, which is expressed by the proposition. But now it is hard to see why this should be the case unless what is doing the real representational job is the thought, i.e., a mental picture, and the proposition is a picture only in a vicarious, derivative way. The intentionality of language piggybacks on the intentionality of the mental.
This very familiar, Lockean view has sometimes been ascribed to the Tractatus. 15 But there is good reason to resist such a reading. The Tractatus contains no explicit distinction between first-rate and second-rate pictures. Thoughts and propositions are simply described as (logical) pictures, and have as such equal status. Given their shared pictorial nature, it is unclear why a mental picture should be able to do something that a linguistic picture cannot do. In fact, if the early Wittgenstein is committed to any priority between language and thought, he seems to give priority -even though only of a heuristic sort -to linguistic pictures. As emerges in the letter to Russell we are discussing, Wittgenstein conceives of mental pictures on the model of linguistic pictures: "[A Gedanke consists] of psychical constituents that have the same sort of relation to reality as words." That is, a thought consists of psychical elements that play the same role of words. The point is conveyed even more explicitly in a 1916 entry of the pre-Tractarian Notebooks: Now it is becoming clear why I thought that thinking and speaking were the same. For thinking is really a kind of language [eine Art Sprache]. For a thought […] just is a kind of proposition [eine Art Satz]. (Wittgenstein, 1979, p. 82, modified translation) This does not mean that a thought is a kind of proposition in the way German and English propositions are kinds of propositions, but in the sense that it is like a proposition (a proposition of sorts, one might say) in all the essential respects, i.e., in all the respects that determine its representational properties. 16 If there isn't necessarily a mental picture behind every linguistic picture, we need a different reading of the claim that propositions express thoughts -one that does not construe the idea of expression as a correlation between something inner and something outer. I suggest that Ramsey points us in the right direction: As to the relation between a proposition and a thought Mr Wittgenstein is rather obscure; but I think his meaning is that a thought is a type whose tokens have in common a certain sense, and include the tokens of the corresponding proposition, but include also other non-verbal tokens; these, however, are not relevantly different from the verbal ones, so that it is sufficient to consider the latter. (Ramsey, 1923, p. 274) The idea I want to take from this passage is that propositions are thoughts. They are thoughts taking a linguistic, sensibly perceptible form. Not all thoughts must take such a form: some may take a psychic, non-linguistic form. A proposition "expresses" a thought in the sense that it embodies or instantiates it, where the instantiation is sensibly perceptible. 17 What all the expressions of the same thought have in common, be they mental or linguistic, is that they are only-logical pictures with the same sense. 18 This leaves us with two difficulties. If a thought may take the form of a proposition, why does Wittgenstein deny, in the letter to Russell, that it consists of words? What it should have said is that it may or may not consist of words. And if a thought does not have to take the form of a proposition, why does the Tractatus say that "[t]he thought is the significant proposition" (4)? Shouldn't he have said that some thoughts are significant propositions? arguments against the priority-of-thought reading, see Winch (1987), pp. 3-17, andMcGuinness (2002), pp. 82-102. 17 One may object that "to express" simply does not mean "to instantiate (in a sensibly perceptible manner)". It means to "press something out", either literally or metaphorically, and in either case there must be something inner and something outer. But consider this use of the term: "This painting is a good expression of impressionist art". It is a good expression of impressionist art in the sense that it is a good instance of it. I suggest that the Tractarian notion of expression should be construed along similar lines. 18 Frascolla (2011, chapter 2) proposes a similar reading, but goes further than I do in maintaining that the tokens of a thought include all the pictures that express the same sense, whether or not they are only-logical pictures.
I propose to solve the first difficulty by holding that the early Wittgenstein uses the term "thought" in (at least) two senses. When he says in the Tractatus that a proposition expresses a thought, the term is used in a wide sense that applies to all only-logical pictures. By contrast, in the letter to Russell the term is used in a narrow sense that applies only to mental only-logical pictures. With regard to the second difficulty, I suggest that "The thought is the significant proposition" should be read as: "The thought (= the only-logical picture) is, in the clearest case, the significant proposition". Not all thoughts consist of words; but what renders any thought an only-logical picture of reality is what can be seen, and can be seen most clearly, in propositions. That is why Wittgenstein feels entitled, in the rest of the Tractatus, to focus only on propositions.
Summing up, I have argued that the primary truth-bearers, for the Tractatus, are pictures. Thoughts are a proper subset of pictures, and propositions are in turn a proper subset of thoughts. A thought is an only-logical picture, and a proposition is a thought taking a linguistic form. There are pictures that are not thoughts, such as spatial or chromatic pictures; and there are thoughts that are not propositions, such as mental only-logical pictures.
So far I have brought out two differences between the Tractatus and Frege. First, for the Tractatus the primary truth-bearers can be (though need not be) sensibly perceptible: they can consist of spoken or written words. Secondly, the primary truth-bearers are not objects, but facts ("The picture is a fact", 2.141). It is a central doctrine of the Tractatus, and a central criticism of Frege, that objects and facts differ categorically. Objects can only be named (3.221), and facts can only be described (3.114). The distinction is as deep and uncompromising as Frege's own distinction between objects and concepts. For Frege, there is no such thing as using a proper name to refer to a concept: the only way to refer to a concept is to use an expression predicatively. Strictly speaking, the term "refer" is itself categorically ambiguous when applied to proper names and concept-expressions: in the sense in which we may refer to objects, we may not refer to concepts. Similarly, for the Tractatus, there is no such thing as using a name to refer to a fact. The only way to think or speak of a fact is to describe it by means of another fact, i.e., to picture it. In the sense in which we may think or speak of objects, we may not think or speak of facts. This is enough to show that the Tractatus cannot accept Frege's relational analysis of indirect speech. And in fact, as we are going to see below, it explicitly rejects any such analysis. But before we get to that point, I want to show that the primary truth-bearers, for the Tractatus, are not mind-independent.

Pictures as facts-in-use
We saw that a Tractarian picture is a fact. But it is not just a fact, for there are many facts that don't represent anything. A picture is a fact together with the "representing relation" (2.1513), i.e., the coordination of the elements of the picture with the elements of the possible situation it represents (2.1514). When the Tractatus introduces the notion of a proposition, it makes clear that standing in a representing relation to reality is a matter of being used in a certain way. The Tractatus distinguishes between the "proposition" and the "propositional sign". A propositional sign is a fact: the fact that its elements, namely linguistic signs, stand to one another in a certain way (3.11-3.14); and a proposition is an "applied" propositional sign (3.5-4), that is a sign that is "used" as a projection of a possible situation (3.11). It seems clear that what is said about propositions is meant to apply to pictures more generally. A picture is a fact that is used as a projection of a possible situation. If this is correct, Tractarian truth-bearers essentially involve use, and thus the act of some thinker or speaker. At the same time, they essentially involve an instrument of representation, namely a fact (either mental or linguistic) that is put to a picturing use. This conclusion is supported by at least two other pieces of textual evidence. First, the notion of a picture is introduced in the Tractatus with reference to picturing agents: 2.1 We make to ourselves pictures of facts.
A picture is something we make. There is no picture without an act and someone performing the act. Secondly, when the Tractatus moves on to consider the case of propositions, it draws a distinction between "signs" and "symbols". A symbol is "everything -essential for the sense of a proposition -that propositions may have in common with one another" (3.31), and propositions are limiting cases of symbols (3.313). A sign is "what is sensibly perceptible in the symbol" (3.2). A crucial feature of the distinction is that "the same sign […] can be common to different symbols": the sign "is", for example, may symbolize now as the copula, now as the sign of existence, now as the sign of identity (3.321-3.323). In order to "recognize the symbol in the sign", we are told, "we must consider its significant use" (3.326). For instance, in order to recognize which symbol the sign "is" is on a particular occasion, we need to look at how it is used, on that occasion, to contribute to the expression of a propositional sense. So symbols, which include propositions, are signs-in-use. Now, if the bearers of truth and falsity are not mind-independent objects, what is it to think or say that something is the case? And what is it to report, in speech or thought, that someone thinks or says that something is case? In other words, how should we analyze propositions of the form "S thinks/says that p"? The Tractarian discussion of this topic (5.541-5.5422) is almost oracular in its brevity and has challenged generations of commentators: 5.542 But it is clear that "A believes that p", "A thinks p", "A says p" are of the form "'p' says p": and here we have no co-ordination of a fact and an object, but a co-ordination of facts by means of a co-ordination of their objects.
Whatever the Tractarian theory of indirect speech amounts to, one thing is clear: it rejects the relational analysis. Wittgenstein states his target in the immediately preceding remark: 5.541 [I]n certain propositional forms of psychology, like "A thinks, that p is the case", or "A thinks p", etc.
[…] it appears superficially as if the proposition p stood to the object A in a kind of relation (And in modern epistemology [Russell, Moore, etc.] those propositions have been conceived in this way.) Wittgenstein mentions Russell and Moore, but if what I argued above is correct, he could have mentioned Frege as well. Propositions of the superficial form "A thinks/says that p" do not assert that an object, the thinking or speaking subject, stands in relation to another object, the proposition that p.
And they do not even assert that the thinking or judging subject, construed as an object, stands in relation to a proposition, construed as a fact. Rather, they "coordinate facts".
But what could that mean? If what has sense, and is true or false, is a fact-in-use, thinking or saying that p should consist in using a fact in a certain way. By the same token, reporting what someone else has thought or said should consist in describing a certain fact (or at least its general structure, if we are not interested in reporting the exact words or "psychic constituents" that were used), and then specifying the way it was used. The question is how such an accomplishment could be thought to have the form "p' says that p". Answering this question in detail, and doing so in a manner that fits what else is going in the Tractatus, would take heavyweight exegetical work. Here I only want to suggest that it might not be impossible to tell a plausible story along the following lines: The "'p'" in Wittgenstein's formula stands for the description of a fact, and the "says that p" stands for a specification of how the fact has been used. Of course, a great deal turns on how such a "specification" is to be construed, since it can't be anything like the ascription of a property to an object. 19 Back to the three platitudes Frege would likely discard the Tractarian account of truth-bearers on the ground that it can't vindicate the three platitudes from which I started. But is it so?
Let's begin with Repeatability and Shareability. Some contemporary advocates of the act-theoretic approach, such as Scott Soames and Peter Hanks, vindicate these platitudes by means of a distinction between acttypes and act-tokens. To think or say the same thing is a matter of producing tokens of the same act-type. I suggest that the Tractatus can adopt the same strategy. A person can think or say the same thing on different occasions, and different people can think or say the same thing, for the same reason for which a person can perform the same action on different occasions, and different people can perform the same action. Actions are obviously agentdependent: there is no action without an agent. Just as Fregean "ideas" don't go around the world without somebody who "has" them, so actions don't go around the world without somebody who "does" them. But this does not mean that actions cannot be repeated and performed by different people. I can now take a leap, and I can do it again, and you can do the same thing. This does not mean that there is an agent-independent object out there, "the leap", to which we both stand in a relation of "taking". To perform the same action is a matter of tokening the same action-type. Similarly, the Tractatus can maintain that thinking or saying the same thing is a matter of using appropriate facts in the same sort of way. The facts must have, in their fully analyzed form, the same "logical multiplicity" (4.04), that is the same number of simple elements with the same combinatorial possibilities; and such facts must be put to the same picturing use, that is, projected onto the same possible situation. If we take appropriate facts and use them in the appropriate sort of way, we think or say the same thing -just as we can do the same thing, say hammering, if we take an appropriate tool and use it in the appropriate sort of way. Now, what about the platitude I called Objectivity? If what is true or false are facts-in-use, how is it possible that it was true, say, that 2 plus 2 equals 4 well before anybody existed to use any fact at all? Some contemporary act-theoretic accounts of truth-bearers, such as Peter Hanks' (2015, p. 27), seek to vindicate this platitude by positing act-types that exist and are true or false independently of the actual existence of any of their tokens. It is questionable, however, that this reification of act-types is compatible with the claim that the primary truth-bearers are concrete, historical acts, and the proposal threatens to collapse into a version of the view that the primary truth-bearers are mind-independent entities (Reiland, 2017). Moreover, the idea that reified act-types are true or false is incompatible with the Tractarian insistence that only facts, and not objects, can be true or false. Finally, once we have introduced reified act-types as bearers of truth and falsity, it is hard to resist a relational analysis of indirect speech. If thinking/saying that p is tokening an act-type, and the act-type is a self-standing entity that exists independently of its actual tokens, then it seems that "tokening" an act-type is a matter of standing in some sort of relation to an object. This is in fact Hanks' account (2015, p. 28). But as we have seen, it is not an account that the Tractatus would accept.
There is another strategy that is open to the Tractatus for vindicating Objectivity. The Tractatus may adopt a deflationist analysis of the occurrence of the truth-predicate in sentences of the form "It is true that p". On this view, the sentence "It is true that p" says the same as "p". But if this is correct, when we say that it was true that 2 plus 2 equals 4 before anybody thought or said it, we just mean that 2+2 equaled 4 before anybody thought or said it. And this, of course, does not require the existence of any thinking or speaking subject. 20 Dummett, M. (1986 The problem Frege, in his "Introduction" to the The Foundations of Arithmetic (FA) (1959), claimed that violating the context principle (CP) leads almost inevitably to a recognition of subjective ideas as the meanings of words -that is, to psychologism in logic. In the present article, my aim will be to show that accepting CP almost inevitably leads to a rejection of the project of giving a completely general explanation of linguistic meaning. What I have in mind is that it is difficult to reconcile CP with any version of the project of giving such an explanation of meaning in a manner that does not appeal to semantic terms: i.e., any version of the project of constructing a full-blooded theory of meaning. The full-bloodedness requirement, as formulated by Dummett (1987Dummett ( , 1993a, can be fulfilled in many ways. At the same time, it should be noted that although Dummett (1993c) was a critic of psychologism, psychologistic explanations of meaning are not actually ruled out by this requirement. So my goal will be to show that accepting CP obliges one to question the range and variety of positions available as regards how one should explain the meaning of linguistic expressions. One of these positions is psychologism, conceived as a standpoint which explains the meaning of linguistic expressions in terms of subjective associations.
I begin with a short characterization of CP. However, my discussion will not focus on historical questions or on presenting alternative interpretations of CP. Instead, I shall outline the reading of CP which I myself embrace, and make just a few remarks in connection with certain questions that have been raised by other commentators. Then I shall briefly characterize the difference between the idea of explaining the meanings of linguistic expressions from outside of any language (the idea of a full-blooded theory of meaning), and that of doing so from inside of a language (the idea of a modest theory of meaning). Then I move on to the main point of my article, arguing that it is difficult to render the consequences of CP compatible with the idea of an explanation of meaning external to all linguistic content. I end with some short remarks on psychologism and the possible sources of the fullbloodedness requirement.

The context principle a. Presentation of the principle
The context principle was formulated by Frege in FA. 1 It appears in the "Introduction" and, among others, in paragraphs 60 and 106: […] never to ask for the meaning of a word in isolation, but only in the context of a proposition (1959, p. X).
Only in a proposition have the words really a meaning. It may be that mental pictures float before us all the while, but these need not correspond to the logical elements in the judgment. It is enough if the proposition taken as a whole has a sense; it is this that confers on its parts also their content (1959, §60).
[…] we must never try to define [explain] the meaning of a word in isolation, but only as it is used in the context of a proposition […] (1959, §106).
As Frege did not, in FA, distinguish Sinn from Bedeutung, one might wonder whether CP should be taken as applying to both of them, or only to Sinn, or only to Bedeutung. Below, I shall treat it as applying to sense (Sinn), and shall not seek to resolve the question of whether it also applies to reference (Bedeutung). According to CP, words do not have any meaning outside of the context of meaningful sentences. (I shall hereon use the terms 'sense' and 'meaning' interchangeably in this article.) Their meaning is determined in the context of a sentence. Furthermore, there is a suggestion in the part of Section 60 quoted above that the meanings of words in a given sentence are logical parts of the judgment corresponding to that sentence. So, in order to determine the meaning of a word in a sentence, one must determine its logical role in the sentence. (I shall be using the term 'logical role' in its wider sense here, according to which the logical role of an expression includes both its syntactic and its semantic features). 2 Moreover, 2 How do I understand the logical role of an expression? Roughly speaking, the logical role of a given expression is fixed, first, by that how its syntactic category determines and is determined by the syntactic category of more complex expressions of which the given expression is part, and, second, by that how its semantic value determines and is determined by more complex expressions, and not only by such expressions of which it is part. This interdependence of the semantic value of a given expression and other expressions shows that an adequate conception of the logical role of the given expression requires giving a description of inferential relations, which hold between various sentences -both the sentences in which the given expression occurs and certain other sentences. Let us begin with syntactic questions. The logical role of the word "John" in the expression "John is brave", which is determined by the fact that under the assumption that syntactic categories of the two expressions among the following three, "John", "is brave", "John is brave" are given the syntactic category of the third expression, is fixed. If one assumes, for example, that "John is brave" is a sentence, "is brave" is a predicate, then "John" must be a name. Now, let us turn to semantic questions. The logical role of the name "John" in the sentence "John is brave" is determined either by the fact that the according to Frege, it is only in virtue of the fact that the whole sentence is meaningful that a sense can be ascribed to its parts. As Frege was to put it much later on: "So I do not begin with concepts and put them together to form a thought or judgment; I come by the parts of a thought by analyzing the thought" (1979, p. 253).

b. Some substantive and interpretative problems pertaining to CP
The first problem with the principle thus formulated is this: it seems to be too strong, and for that reason not credible. Why? Because we often do use semantic value of this name determines the semantic value of this sentence (this case is, of course, more common) or by the fact that the semantic value of this sentence -when it is fixed -determines the semantic value of this name (in such cases more information is needed to decide whether the name "John" functions like a proper name or like a description of the type "a person whose name is "John" and who is brave"). When one describes the logical role of a given expression, one cannot, of course, confine to showing how the given expression functions in only one more complex expression. As I have already mentioned, that is the reason why the logical role is determined by inferential relations which hold between various sentences. So, the logical role of the name "John" is determined, among other things, by the fact that the sentences "There is someone who ate a big breakfast", "A big breakfast was eaten by John", can be inferred from the sentence "John ate a big breakfast". An analogous relation holds in the case of expressions belonging to other syntactic categories, for example, "is red", "and". For example, the fact that one can infer the following sentences "This car is coloured", "There is a property such that this car has this property" ("There is an F such that this car is an F") from the sentence "This car is red", determines in some -of course, not very specific -way the logical role of the predicate "is red". Whereas the logical role of the word "and" is determined in natural languages in a significant way by introductory and eliminatory rules that allow one to infer the sentence "John is clever" from the sentence "John is brave and clever" and the sentence "John's car is red and cheap" from the sentences "John's car is red" and "John's car is cheap". (What is the nature of the relation between rules of use of logical constants and their meanings is by some philosophers perceived as a controversial question. However, I will not discuss this issue here [cf. Prior, 1961;Diamond, 2002]). However, it is worth adding that in natural languages, the order of clauses with a conjunction is in some cases essential for the sense of the whole sentence, and for that reason the principle that "p and q" entails "q and p" is not valid without exception in natural languages ("John went out of the house and ate breakfast" does not entail "John ate breakfast and went out of the house"). The above remarks are not, of course, a full explanation of the concept of the logical role, but I think they are sufficient to understand what I mean by the term "the logical role of an expression". single words or incomplete sentences in a conversation, and these expressions are undoubtedly meaningful. This problem has been discussed by, among others, Glock and Bronzo (Glock, 2004;Bronzo, 2011). It can be resolved, roughly speaking, in the following way: uses of words outside of the context of a sentence are derivative: the meaning of a word used outside of the context of a sentence is derived from the meaning of a word used within the context of certain meaningful sentence (cf. Bronzo, 2011). How, exactly, one should understand this derivative character is a question I shall not address here.
It is worth pointing out that there are languages in which it is possible to construct sentences consisting of only one word (for example, my first language is such a language) but that does not imply that certain single words are identical to certain sentences. One of the visual indicators of the fact that a certain expression is a full sentence is that a full stop, a question mark, or an exclamation mark occurs at the end of a string of words (one word is, of course, also a string of words) of which this expression consists (i.e. a full sentence is not identical with any string of words -a full stop, a question mark, or an exclamation mark must be added to an appropriate string of words in order to obtain a full sentence). It is worth adding that such a full sentence can, of course, consist of other, simpler sentences, and those simpler sentences (i.e., the clauses the full sentence is made of) are the strings of words that will become full sentences when completed with appropriate punctuation marks. Full sentences are used to assert that something is the case, ask whether something is the case, order that something be the case, and so on; whereas words are used to construct sentences. This view is often criticized. One points out that in many cases, uses of single words -and not such words that form one-word sentences -are speech acts, such as assertions, questions, commands (Stainton, 2000). Examples used to support this thesis are such strings of words that, from a grammatical standpoint, are not correctly-constructed sentences. However, one cannot say that uttering these strings of words is merely an uttering of words and nothing more because at the end of these strings occur full stops, question marks, or exclamation marks. This fact indicates that they serve to express assertions, questions, or commands. On the other hand, it is justified to say that one utters nothing but words when, for example, a pupil repeats certain words after the teacher or rewrites them from the board. This shows that there is even a syntactic difference between expressions that are used to express assertions, questions, or commands, and words. The existence of this difference does not, of course, prove that strings of words used to express assertions or questions are sentences. What one can undoubtedly acknowledge is that such expressions are used in a similar way to sentences, but their surface form is different from the surface form of sentences. It also seems that users of natural languages understand such expressions because they understand sentences that consist of words occurring in these expressions. To sum up, sentences and expressions that serve the same purpose as sentences are used to perform illocutionary acts, such as assertions, questions, commands, etc. These illocutionary acts have fulfillment conditions, whereas words qua words are used to construct sentences or such expressions which function as sentences; other cases of word use, such as a mere utterance or the writing of a word or words do not have fulfillment conditions.
The second problem concerns the possibility of reconciling CP with the principle of compositionality of sense. If these two principles are accepted, the following question then arises: does this not lead to a vicious circle in explaining the sense of linguistic expressions (cf. Bar-Elli, 1997)? If, according to the principle of compositionality, the sense of a whole sentence depends on the senses of its parts and the way they are combined, and if, according to CP, the senses of parts of a sentence are determined by the sense of the whole sentence, then to explain the sense of a whole sentence one must already know it, as an explanation of the sense of the whole sentence requires an appeal to the senses of its parts, but according to CP their senses are determined by the sense of the whole sentence. One can try to solve this problem in different ways. Firstly, one can abandon one of these principles. Secondly, one can reinterpret the principles in such a way as to avoid the vicious circle. However, neither of these strategies seems credible.
There are many different interpretations of CP. However, I shall just touch on two points pertaining to some popular interpretations of the principle. First, I shall briefly discuss Dummett's remark concerning Quine's formulation of CP, and then I shall present the distinction made by Glock between weak and strict versions of CP (Glock, 2004). Quine, in Two Dogmas of Empiricism (1961), claimed that, according to Frege, the unit of sense is a sentence, not a word, but in Dummett's opinion, Quine's formulation of CP is either a truism or an absurdity (Dummett, 1973, p. 3). If Quine's statement is understood as meaning that one can only say something by means of a sentence, then it is the former. However, if Quine's statement implies that single words, like single letters or isolated syllables, do not have any meaning, then it is absurd. As Diamond rightly points out, Quine's statement can and should be understood in a different way (Diamond, 1991). CP says that words which occur in sentences have sense in virtue of the fact that in each case the entire sentence has sense. That is why the concept of identity of sense between expressions can be applied only to expressions occurring in sentences. The question of identity of sense between expressions taken as occurring outside of the context of a sentence is meaningless. Such a reading of Frege's CP is undoubtedly supported by the following quotations: It is enough if the proposition taken as a whole has a sense; it is this that confers on its parts also their content (1959, §60).
So I do not begin with concepts and put them together to form a thought or judgment; I come by the parts of a thought by analyzing the thought (1979, p. 253).
Glock, in All kinds of nonsense, distinguishes two versions of CP. The first, strict version, says: "A word (name) has meaning only in the context of a proposition" (2004, p. 225), while according to the second, weak version, words have meaning because they can be used in sentences: The kernel of truth in contextualism is that the meaning of a word is determined by how it can be used within sentences. But it does not follow that the word has meaning only in the context of a sentence. On the contrary, it is the individual word which has such a use (2004). However, it is obvious for Glock that Frege (in FA) accepted the strict version of CP. According to Glock there is no doubt that the strict version of CP is wrong. The most fundamental reasons for rejecting it are to be found in the existence of counterexamples (occurrences of words in dictionaries, lists of words, etc.) and the impossibility of reconciling CP with the principle of compositionality of sense.
On the basis of the above remarks, one can draw the following conclusion: if one accepts CP (interpreted in the Fregean way), then it is not possible to explain the sense of a sentence in terms of senses of the words forming it already given prior to the sense of the sentence itself, and neither is it possible to claim that recognition of the sense of a sentence is based on prior recognition of the senses of the words forming it. This conclusion directs our attention once again to the question of whether it is possible to reconcile CP with the principle of compositionality, while the conclusion imposes quite substantial constraints on possible solutions to that problem. I will start with an attempt to answer the question of whether Dummett's sketch of a solution to this problem fulfills this condition: namely, that one should not treat meanings of words as prior to meanings of sentences. The following quotation presents Dummett's view: (…) in the order of explanation the sense of a sentence is primary, but in the order of recognition the sense of a word is primary (1973, p. 4).
One can say that Dummett solves the problem of how to reconcile the two principles through a reinterpretation of CP. According to his solution, the principle of compositionality is primary in the order of recognition, whereas CP is primary in the order of explanation. But how should we understand the issue of priority here as it relates to the order of recognition? Dummett explicates this point in the following terms: We thus derive our knowledge of the sense of any given sentence from our previous [my underlining] knowledge of the senses of the words that compose it, together with our observation of the way in which they are combined in that sentence (1973, p. 4).

The conception presented above of what is involved in understand-
ing the sense of a sentence assumes that knowledge of the senses of words forming a sentence is prior to knowledge of the sense of the whole sentence. This, in turn, means that one must grasp the meanings of words independently of the meanings of sentences. So does this imply that one must grasp the meanings of words, so to say, in isolation? Dummett himself seems to repudiate such a conception, in that he claims that "our understanding of those words consists in our grasp of the way in which they may figure in sentences in general" (1973, p. 5). No matter whether Dummett's conception is coherent or not (i.e., whether or not a prior grasp of the sense of the words requires that they be grasped, so to say, in isolation), this is by no means compatible with the condition to the effect that our understanding of the words forming a sentence is not prior to our understanding of that sentence itself. It is worth noting that the version of CP endorsed by Glock is also not compatible with this condition.
It therefore seems that the only possible option for someone who accepts CP (in its proper sense) is to acknowledge that knowing the sense of a whole sentence is indeed prior to knowing the sense of its parts. Even so, there are many objections to this kind of solution. I will not discuss them systematically here, but just point out the most important ones. First of all, if the meaning of sentences were to be prior to the meaning of words in the sort of way that would entail that we would be able to understand sentences without recognizing the words that make them up, then the existence of logical connections between sentences would be a complete mystery. Secondly, the conception of the priority of the sense of sentences to the sense of words would seem to be compatible with the possibility of treating sentences as completely devoid of semantic structure: if the sense of a sentence were prior to the sense of the words, one would never need to grasp the sentence's structure in order to be able to grasp the sense of the sentence itself. This, then, obscures the difference between sentences and simple names. Thirdly, acknowledging the priority of sentences to words in such terms makes it impossible to explain the fact that we are able to understand indefinitely many new sentences. That argument has, of course, been put forward on many occasions in the context of discussions on the relationship between CP and the principle of compositionality (Dummett, 1973, p. 4).
I myself think that these two principles are compatible. In essence, I agree with Bronzo's proposal for how one ought to construe the relationship between the principles (2011). I will not present his considerations in any detail here, though. Rather, I will just confine myself to pointing out that in his opinion one should reject the conviction that either the sense of a sentence must be prior to the senses of its parts, or the senses of words forming a sentence must be prior to the sense of that very sentence. The recognition that these two principles are in fact complementary is what offers a genuine solution to the problem. So grasping the sense of a sentence requires grasping the sense of its parts, and grasping the sense of parts of a sentence requires grasping the sense of that whole sentence. But doesn't such a solution result in a circular explanation of meaning? And is it not therefore entirely lacking in credibility? I shall not seek to answer these questions at this particular juncture, but instead will return to them in subsequent parts of this article.

c. CP and moderate holism
How should one set about explicating the meaning of a complete sentence? Can one do so independently of any explanation of the meaning of other sentences -in isolation, so to speak? It seems that if one attempted to explain the meaning of a given sentence without making any appeal whatsoever to the sense of any others, one would inevitably fall prey to psychologism, because such explanations of the meaning of a sentence would not then be able to take into account its logical relations with others, and for that reason would probably involve an appeal to subjective associations. As such, it seems that this would also make it impossible to see the logical connections between sentences, and so would violate the principle of compositionality of sense. Thus, any such extremely narrow interpretation of CP is completely implausible. Moreover, the following understanding of CP seems quite natural: the meaning of a word should be explained not in the context of one sentence, but in the context of many sentences. If one were to give, for every context in which a word occurs, a separate explanation of the meaning of this word, it would imply that in fact there were no words (one could say that there would be no semantic difference between such words and mere clusters of letters or syllables). So according to CP, in order to explain the meaning of a word, one must know the meaning of a whole body of sentences in which this word occurs (Davidson, 1984) -sentences that are logically interconnected. For example, the explanation of the meaning of the English word "cat" requires knowledge of the meaning of sentences like "this is a cat", "this is not a cat", "there is a cat here", "there is no cat here", "the cat stands", "the cat lies", "the cat runs", "cats are animals", "cats are born", "cats eat", "cats die". 3 Of course, the question of how many sentences of this kind one has to understand in order to know the meaning of the word "cat" is not only controversial but in some sense unanswerable. However, there can be no doubt that knowing the meaning of the word requires that one understands at least some sentences of this sort.
Even so, such a conception of knowing the meaning of a word can be questioned. First, it seems to lead to an unacceptable holism: that is, to a view according to which knowledge of the meaning of a word requires full mastery of an entire language. Why? To explain this, let us suppose that knowing the meaning of the word "cat" requires that we understand the above-mentioned sentences. Of course, understanding these sentences itself requires us to understand their parts: namely, amongst others, the words "it", "not", "is", "here", "no", "to run", "animal", and so on. In turn, understanding these words requires understanding other sentences in which they occur. These other sentences will be composed of yet other words, the understanding of which requires understanding some other set of sentences, and so on. So, understanding the word "cat" would then require mastery of a whole language. However, it is worth noting that this statement is only really absurd if taken to mean that knowledge of the meaning of the word "cat" requires a full understanding of the entire English language. As Dummett rightly points out, the Wittgensteinian dictum that "to understand a sentence is to understand a language" (Wittgenstein, 1974, §199) can be interpreted in another much more reasonable way: understanding a word requires mastery of just some part of a natural language that can itself be recognized as making up a complete, though certainly also a very primitive, language (cf. Dummett, 1993d, p. 222).
The second reason why it may seem that the conception of understanding a word that requires us to understand a certain set of sentences may be criticized is that it does not furnish a criterion that would allow us to decide in every case whether a given sentence belongs to this set or not. This could be an objection, but only if the concept of knowing the meaning of a linguistic expression (i.e., possessing the requisite linguistic understanding) is a sharp one. What this impossibility shows, then, is that concepts such as knowledge of meaning, understanding, etc., are not like that. Some people undoubtedly do know the meaning of a certain word because they are able to deploy a sufficiently large number of sentences in which that word occurs (of course, that ability consists not only in a disposition to use these sentences in appropriate (non-linguistic) circumstances, but also in the ability to recognize the logical connections between them), and other people undoubtedly do not understand the word in question, because they cannot make use of any sentence in which it occurs. However, it happens that there are people who can only use a very limited number of sentences in which the word occurs. Moreover, even this competence is highly limited, as such people cannot use these sentences in many circumstances in which a competent language-user would be able to use them, and cannot see many fundamental logical connections between them. The obvious instance of such a person is a young child who distinguishes cats (such as are typical for Europe) from dogs, but who, if asked, "Is it a cat?" when presented with a Siamese or a Persian cat, does not answer, or says, "I don't know". Such a young child might well accept not only sentences like "cats mew", "cats drink milk", but also ones like "my plush cat drinks milk", or "my plush cat was born once". Moreover, it is quite plausible that such a young child will not understand, and therefore will not accept, the following inference: because all cats are animals and, in fact, no plush cats are animals, no plush cats are cats. No matter how one may seek to account for the details of the above example, it shows clearly that the concepts of meaning and linguistic understanding are vague and gradable.
The answers just presented to these two objections -namely, the objection that embracing CP must lead to a radical and unacceptable version of holism, and that it is categorically false to suppose that the concepts of meaning and understanding are vague and gradable -allow one to hold that a moderately holistic approach to meaning is not after all beyond the bounds of plausibility. 4 The difference between explaining meaning from the outside and from the inside with respect to conceptual/linguistic content We find the most radical kind of skepticism towards the idea of explaining meaning from the outside in respect of linguistic content exhibited in the following remark by Wittgenstein: The limit of language manifests itself in the impossibility of describing the fact that corresponds to (is the translation of) a sentence without simply repeating the sentence (1998, p. 13).
Wittgenstein pointed out that if one wanted to explain what is asserted by means of a certain sentence, one would have to use the same sentence again. Of course, Wittgenstein's remark is not entirely convincing when taken literally, because in many cases where one wants to capture the fact corresponding to a sentence one can actually just employ a paraphrase of that sentence. So, Wittgenstein's remark would be adequate, were it to be supplemented with this qualification. However, Wittgenstein's formulation has a certain didactic value anyway, in that it serves to emphasize the trivial character of explanations of meaning.
What does it mean to say that such an approach to the explanation of linguistic meaning is trivial? Primarily, it means that on this approach, it is not possible to explain meaning in non-semantic terms: i.e., that any attempt to do so is doomed to failure. At the same time, though, the triviality just adverted to need not necessarily entail that it is impossible to reveal important relations between the notion of meaning and other semantic and non-semantic notions. The most important contemporary philosophers who endorse such an approach are John McDowell and Barry Stroud (McDowell, 1998a, 1998bStroud, 2000aStroud, , 2000bStroud, , 2000c. I will call their approach "modest". According to this way of looking at things, it is not possible to explain meaning without appealing in some way or other to the notion of linguistic or conceptual content. (Both adherents and critics of this approach regard the concept of conceptual content as a semantic notion in the broad sense of this term.) Thus: What adherents of modesty deny, then, is the feasibility of this sort of account of the practice of speaking a language: an account given from outside the thoughts expressible in the language, and indeed from outside the very idea of expressing thoughts, but nevertheless such as to display what makes the behavioural repertoire it deals with a case of speaking a language. (McDowell, 1998 b, pp. 111-112).
Similarly, statements about what expressions mean […] are not equivalent or reducible to non-semantical or non-intentional descriptions of what goes on when certain sounds or marks are made. (Stroud, 2000b, p. 187) By way of contrast, I shall call the alternative approach to explanations of the meaning of linguistic expressions "full-blooded". According to this view, both linguistic and conceptual content should be explained in terms of some more basic concepts. The terms "modest" and "full-blooded" were used in this context for the first time by Dummett, who originally described his own approach as full-blooded, and both Davidson's and McDowell's as modest (Dummett, 1993b). (He later changed his mind as regards Davidson's project (cf. Dummett, 1993b, Appendix)). It should be noted that on such a construal of what this alternative amounts to, not only Dummett's conception, but also, for example, that of Fodor (cf. 1992, p. 32;2008, p. 203), counts as full-blooded. It should be added that Dummett's conception is not as reductive in spirit as Fodor's, in that it emphasizes the rational character of language use and does not aim to formulate predictions about what speech acts will occur (Dummett, 1987).
At the same time, I would like readers to be alerted to the fact that I shall not pay much attention to the question of whether I attach precisely the same significance to the terms "modest" and "full-blooded" as McDowell and Dummett. This is because I shall be using these terms only as convenient labels for the approaches I myself am concerned to characterize in this part of my article.
Acceptance of the modest approach to meaning implies that adequate explanations of the meaning of any expression will involve such terms as "expresses thought", "states that", "expresses truth", "is true" or even "means that". Of course, some modest explanations exhibit a manifestly circular character and, for that reason, are completely uninformative. Moreover, it seems that such explanations as are not explicitly circular are ultimately based on ones that are so. For example, the meaning of the sentence "John is an eccentric person" can be explained in the following way: this sentence means that John is a person who is different from other people and whose behaviour is considered by other people to be strange and unusual. However, such an explanation -if one's aim is to give the meaning of every expression belonging to a given language (in this case English) -requires further explanations. It is obvious that we will come, sooner or later, to an explanation of the form: the sentence "x is F" means that x is F. An explanation of this form is, of course, manifestly circular because it can be understood only by someone who can use the sentence "x is F": i.e., someone who knows the sense of this sentence and understands the concept of meaning. So how could one avoid this circularity? Basically, one might do so in one or other of two ways: either by rejecting modesty and accepting full-bloodedness, or by formulating explanations in other semantic terms. As regards the latter, acceptance of a truth-conditional conception of meaning (cf. Davidson, 1984) can be treated as an attempt to avoid this kind of circularity without abandoning modesty. The meaning of a sentence of the form "x is F" is then explained by means of a bi-conditional formulated in a metalanguage: e.g., "the sentence 'x is F' is true only if x is F". This, of course, cannot be the whole story where this particular conception of meaning is concerned. But can this conception avoid the objection of circularity? That, of course, will depend on exactly how one understands the term "circularity". If one recognizes as circular any explanation which has to appeal, in some sense, to notions of linguistic or conceptual content, then this conception also ought to be considered circular. Why? Because even if one were to acknowledge, as Davidson did, that the concept of truth is primitive and undefinable (1996), one would not be in a position to deny that it presupposes the concept of meaning. (If one were to opt to use the notion of truth-in-a-language as defined by Tarski to construct a theory of meaning, then its connection with the concept of meaning would be obvious. Tarski's definition assumes, firstly, the concept of an interpretation of a language and, secondly, if a metalanguage does not contain the object-language, the concept of translation.) The use of the concept of truth in explanations of the concept of meaning requires that it be comprehensible itself. So, if the sense of the predicate "is true" is not given by a definition, then it must be given in some other way. And indeed, it is given to us by the way we use this predicate. How do we use it? Firstly, we recognize that it can be applied only to meaningful sentences which express statements. Secondly, we ascribe it to some such sentences, among others, in virtue of their senses. These remarks show that the conception which explains the meaning of sentences in terms of their truth-conditions is, in some sense, circular. However, explanations which are circular need not necessarily be completely idle and uninformative. Such explanations can be called "non-manifestly circular". They can show essential connections within a given set of concepts. If a circle of concepts is not too small, explanations can be revealing, though not, of course, in the sense of all concepts belonging to the set being fully explainable in terms of concepts belonging to its proper subset (cf. Strawson, 1992, pp. 19-20).
The above considerations show that one of the main reasons for accepting the full-blooded approach to meaning is the aim of avoiding circularity of any kind when explaining it. What, though, motivates the modest approach? The most basic consideration here is, I think, the conviction that all full-blooded conceptions of meaning misrepresent the phenomenon of linguistic meaning. According to McDowell and Stroud, one of the absurd consequences of full-blooded conceptions is that if they were to be correct, then the meanings of the expressions we use would always be completely indeterminate (McDowell, 1998b;Stroud, 2000b). Why so? The point is that full-blooded conceptions assume that the meaning of any expression is constituted by certain features of the expression that are describable in nonsemantic terms (e.g. the use of the expression described non-semantically, the association of the expression with a certain class of images in one's mind, a causal connection between the expression and a certain class of things, etc.). However, Kripke's (1982) considerations show that the fact that a given -non-ambiguous -expression has any property which can be described in non-semantic terms is compatible with ascribing to this expression one of infinitely many different -mutually incompatible -meanings. So, if Kripke is right, full-blooded conceptions of meaning seem to support the thesis of complete meaning indeterminacy. This thesis, however, as Stroud rightly points out, leads to the paradoxical conclusion that the sense of this very thesis is itself also indeterminate (2000b). Meanwhile, McDowell notes that Kripke's considerations also entail a quite implausible conception, according to which understanding a language is always a hypothetical matter, in that one can never be sure what someone else's words mean (1998b).
Because my aim here has not been to actually resolve the dispute between them, my presentation of these two alternative approaches to meaning (the modest approach and the full-blooded one) has confined itself to just giving a rough outline sketch of the arguments for both sides.

CP and modesty
The main aim of my article is to show that acceptance of CP almost inevitably leads to rejection of the project of constructing a full-blooded theory of meaning, and thereby also to the recognition that meaning can only be fully explained, if at all, in a modest way, i.e., in terms of semantic notions. So how might one justify such a conviction?
I would like to begin my line of argument by reminding readers that CP excludes the possibility of understanding a word without having knowledge of the meaning of a fairly extensive set of sentences containing that word. Hence, if one's mode of explaining the meaning of linguistic expressions is to respect CP, one's explanations cannot have the following form: first, the meanings of words and the possible modes of their combination are given without, of course, any explicit or implicit appeal to meanings of sentences; then, on the basis of that first step, one is able to give the meanings of sentences composed of these words combined in the previously specified ways. (It should be emphasized that this statement does not rule out explanations of linguistic meaning of the following kind: the Polish word "kot" denotes the class of cats, the Polish expression "X je" means that X eats, etc. Why? Because if the meanings of the English words are known, and are so on the basis of knowledge of the meanings of a quite extensive set of English sentences, then explanations of that kind do not presuppose the possibility of explaining the meanings of all words in all languages without appealing to knowledge of the meanings of a relatively extensive set of sentences belonging to a given language.) Now I turn to my second point. I shall seek to show that the mode of explanation excluded by CP can look attractive from the standpoint of the full-blooded approach. Now according to the full-blooded approach, the meaning of linguistic expressions should be explained in non-semantic terms. It seems that even if one agrees that it is possible to explain meaning in such a way, there are serious doubts as to whether it is possible to explain the meanings of words and the meanings of some sentences containing these words at the same time. Yet one would have to fulfill this condition in order to respect CP. What are the reasons for doubting the possibility of such explanations? I think that confronting highly complicated theories of this type (i.e., full-blooded and holistic) with actual linguistic practice would give no determinate indications as to how such theories might be improved. (The supposition that one can just formulate a completely adequate theory all at once, so to speak, is of course not at all credible.) What I mean by this is that the incompatibility of such theories with linguistic practice can be explained in too many alternative ways that are by no means completely implausible. For example, the fact that such a theory ascribes an understanding of the word "cat" to someone who evidently does not know the meaning of this word can be explained in terms of the thought that the theory mistakenly identifies the meanings of such words as "animal" or "alive" with certain properties (properties of human organisms or certain complicated relational properties), or that the theory mistakenly describes the character of logical relations -and so on. Of course, the above remarks do not prove that a full-blooded theory cannot be holistic; however, they show that accepting even a moderate version of holism calls into question the possibility of constructing a full-blooded theory of meaning. Moreover, it seems that the requirement of full-bloodedness is much more in agreement with semantic atomism. If one wanted to present a full-blooded theory against the background of the assumption that semantic atomism was correct, one would only need to explain the meanings of a basic lexicon in non-semantic terms. That is to say, it would be enough to separately explain the meanings of those words which cannot be defined in terms of the meanings of other words. When a theory that is both full-blooded and atomistic is brought face to face with actual linguistic practice, this does not seem to lead to such difficulties, as in the case of a theory that is full-blooded but holistic. In the case of an atomistic and full-blooded theory, it is easier to pinpoint elements of the theory that run counter to linguistic practice. To sum up the second point of my argument, then, I have tried to show that it is hard to reconcile even a moderately holistic approach to meaning with the requirement of fullbloodedness, and that fulfilling this requirement seems much more plausible against the background of the assumption that semantic atomism is right.
The third point of my argumentation concerns the following question: what character can explanations of meaning that respect CP have? I have already pointed out that they must be at least moderately holistic. Now I will try to show that provided that one accepts a certain (in my view) quite natural generalization of CP, such explanations must also be modest. In order to make my argument more perspicuous, it is worth reminding ourselves what conception of the explanation of the meaning of linguistic expressions (conceived as logical units) is implied by CP. I take CP to imply that the explanation of the meaning of any expression consists in determining what logical (semantic and syntactic) role it plays as a component of other meaningful expressions. It should also be added that Frege, in the "Introduction" to FA, points out that respecting CP protects one from the danger of psychologism, i.e., from, amongst others things, invoking irrelevant factors in explanations of the meanings of linguistic expressions. The juxtaposition of these two remarks may lead to a posing of the following question: should the principle that states that one expression must be explained in the context of other expressions of which it is a part be restricted to just parts of the sentences, or should it also apply to whole sentences themselves? The adherent of the full-blooded approach who also wants to respect CP would claim that the meanings of sentences should be explained in a different way, that is, without any appeal to meanings of other expressions. According to adherents of the full-blooded conception, meanings of sentences should be explained in non-semantic terms (e.g., in sociofunctionalist terms, or with reference to causal relations holding between, so to speak, syntactically-defined states of human brains and the world, etc.). It seems, however, that even if such an approach is in agreement with the letter of CP, it is not compatible with its spirit. The principle was formulated in order to show that the meaning of expressions is determined only by their logical (semantic and syntactic) roles, and not by such irrelevant factors as, for example, subjective associations prompted by the occurrence of expressions. So strictly speaking, all explanations of the meanings of sentences appealing to something other than the logical role of those expressions will be incompatible with the spirit of CP. Full-blooded explanations of the meanings of sentences have that character just by virtue of aiming to explain the semantic function of sentences in non-semantic terms. It also seems that the burden of proof then lies with the adherent of a holistic and full-blooded theory of meaning to show that the meanings of sentences should still then be explained in a completely different way than the meanings of words -i.e. not in terms of their logical role. To sum up, the natural extension of the Fregean principle is the requirement to explain the meaning of any given linguistic expression just through determining its logical role in a language. This generalized version of the context principle -which one could call "Wittgenstein's Principle" -only allows for modest explanations of the meaning of linguistic expressions.

Concluding remarks on psychologism and full-bloodedness
Having argued that accepting CP leads almost inevitably to an adoption of the modest approach to meaning and a rejection of the full-blooded approach, I would like to briefly make two further remarks. The first concerns the relationship between psychologism and the full-blooded approach to meaning, while the second concerns the roots of the idea that we should seek to give a full-blooded explanation of meaning.
Psychologism as regards linguistic meaning can be defined in many ways. Yet no matter how, exactly, one determines the content of this standpoint, it is, broadly speaking, a view according to which the meaning of any expression can be explained in terms of mental objects, acts, states or processes (cf. Dummett, 1973, pp. 637-642;Mohanty, 1997). What is it that is questionable about such a view? Above all, it is the fact that, according to this position, features of expressions which do not play any logical (semantic and syntactic) role in language use are nevertheless treated as relevant, and even essential, for determining the meaning of these expressions. For example, psychologistic explanations of meaning may include images associated with expressions, as if these were something vital to understanding. So the most objectionable aspect of psychologism is its attempt to explain meaning in non-semantic terms. Therefore, it is the full-bloodedness of psychologism that is its most questionable aspect. However, explanations of meaning which appeal to terms referring to mental states need not raise any objections. For instance, explaining the meaning of the sentence "Cats mew" by stating that "'Cats mew' expresses the thought that cats mew" need not be regarded as confusing what is subjective with what is objective. Of course, adherents of standardly construed psychologistic explanations of meaning do not confine themselves to such innocent uses of mental terms 5 in their conceptions of meaning, and that is the reason why these conceptions can be described as full-blooded.
I think that there are two possible main sources for the idea that we should seek to give a full-blooded explanation of linguistic meaning. The first is the aspiration to avoid circularity in explanations, while the second is an acceptance of reductionism with respect to linguistic meaning.
According to those who embrace the full-bloodedness requirement, every attempt to explain linguistic meaning in semantic terms (no matter how the latter are understood) is ultimately flawed, due to the fact of being caught up in a vicious circle. Concepts such as possession of truth-conditions by a sentence or expression of thought by a sentence cannot be understood independently of the concept of meaning, and that is the reason why the meanings of linguistic expressions should be explained in other terms. Thus, it could seem from a logical point of view as though the modest approach to meaning is fundamentally mistaken. My answer to that objection -though somewhat sketchy, I admit -can also be deployed as an answer to the previously mentioned similar objection to Bronzo's proposal for reconciling CP with the principle of compositionality. As I have already claimed, not all kinds of circularity in explanations need be taken to constitute a flaw. Circular explanations can show that there is an essential connection between members of a certain class of concepts. Of course, the circle should be appropriately wide, lest the explanations in question prove completely uninformative (Strawson, 1992). Moreover, it is worth adding that the very attempt to step outside the circle can generate explanations of the concepts and phenomena under consideration that are completely mistaken -or, at least, that can be described as revisionary.
As I said above, the second source of acceptance of full-bloodedness may be the conviction that some version or other of reductionism with regard to linguistic meaning must be right. However, it should be emphasized that Dummett did not himself seek to justify the requirement in that way. Adherents of reductionism assume that irreducible semantic facts would be in some sense queer, 6 and for that reason should be reduced to facts of some more basic kind or other. The above remarks on the possible sources of the full-bloodedness requirement lend support to a certain very general and conditional conclusion: if CP as interpreted here is right, and the line of argument proposed regarding the connection between CP and the full-bloodedness requirement is correct, then the conviction that the semantic aspect of language can be fully explained in certain more basic terms is questionable. Of course, as Stroud rightly points out, such a conclusion is, from a metaphysical point of view, deeply unsatisfactory (Stroud, 2000b, p. 192). However, it should be emphasized that the recognition that it is doubtful whether the requirement of giving an entirely general explanation of meaning could ever be fulfilled is not only a recognition of a certain negative thesis, but also something that permits one to better understand the character of the metaphysical project that aims to explain meaning from, so to say, outside of linguistic content. Of course, adherents of the requirement may well opt to reject the above conclusion, pointing out that "one philosopher's modus ponens is another philosopher's modus tollens" (Putnam, 1992, p. 295). Wittgenstein, L. (1998). Culture and Value. Ed. G.H. von Wright. Oxford: Blackwell.

Abstract
The aim of my article is to show that accepting the context principle (CP) almost inevitably leads to a rejection of the project of giving a completely general explanation of linguistic meaning. I will argue that it is difficult to reconcile CP with any version of the project of giving such an explanation of meaning that does not appeal to semantic terms. I will begin with a short characterization of CP. I will outline the reading of CP which I myself embrace. Then I will briefly characterize the difference between the idea of explaining the meanings of linguistic expressions from outside of any language, and that of doing so from inside of a language. Then I will move on to the main point of my article, arguing that it is difficult to render the consequences of CP compatible with the idea of an explanation of meaning external to all linguistic content. Nearly a hundred years have passed since the publication of Ludwig Wittgenstein's Tractatus Logico-Philosophicus, yet it still provokes heated interpretative disputes. The broad range and sheer number of readings that have emerged, of almost every issue raised by this short book, foster the impression that we, as a community of thinkers, know even less about the author's actual intentions now than half a century ago. However, such a pessimistic conclusion would, I think, be deeply mistaken. Thanks to a more accessible and comprehensive Nachlass, we now have a better sense of Wittgenstein's direction of development as far as his early-period thinking is concerned, as #0# well as with regard to his main sources of inspiration. In a sense, then, the current diversity of interpretative standpoints is just what we would expect, given the evolution in the state of our knowledge. My own starting point, the justification for which lies outside of the scope of the present paper, is the so-called elucidatory interpretation presented by Daniel Hutto and Marie McGinn. They suggest that the content of Wittgenstein's first book should be considered as a collection of clarifications on how to use (or not use) certain concepts, not as a theory of the relation between language and reality. In this, they stand in opposition to both the traditional metaphysical reading as well as the so-called resolute reading offered by Cora Diamond and James Conant (cf. e.g., McGinn, 2006, pp. 1-27). This is compatible with Wittgenstein's general intentions, as laid down explicitly in the Tractatus' remarks on philosophy.
This claim must, however, be accompanied by a reservation: I do not consider the Tractatus to be entirely faithful to its own metaphilosophical stance. The reason for this is that although Wittgenstein radically transformed the conception of logic he had inherited from Russell and Frege, he was not radical enough. The Tractarian idea of logic called for something more than just a set of elucidations of how our language works -it required, at the very least, a minimal metaphysical foundation. 1 Due to this fact, Wittgenstein's early thought must ultimately be regarded as incoherent. 2 In the present paper I shall defend an interpretation of the Tractatus based on the following three theses: 1. The Austrian philosopher's work offers a double-layered vision of language, similar to the vision developed during his brief phenomenological period: the core of each symbolism is a universal structure which to some extent resembles a phenomenological primäre Sprache -i.e., a layer of language standing in an absolutely direct and immediate relation to reality.
1 The notion of minimal metaphysics in the Tractatus may be considered as an elaboration on the intuition of "anti-metaphysical metaphysics of the symbolism" as formulated by Maciej Soin (2001, p. 59). 2 As Michael Kremer has noted, the Tractatus is to a certain extent a transitional work (Kremer, 1997, p. 91, 98, 109). Elsewhere I myself have propounded the view that almost all of Wittgenstein's works, save his very last writings, are in some sense transitional (Gomułka, 2019). 2. This relation to reality must, however, be understood in quite specific terms, given that the so-called ontology of the early Wittgenstein is actually a purely formal construction, entailed by the structure of what we shall refer to as the inner layer of language. (One should consider Heinrich Hertz's Prinzipien der Mechanik as the main source of inspiration in this respect.) 3. It should be recognized that the metaphysical residuum within the early Wittgenstein's thought is a certain minimal form of transcendentalism, according to which language -or strictly speaking its inner layer -performs the function of the transcendental subject for itself. Although each of these three theses may be entertained separately, together they make a consistent whole: at least, that will be the approach I defend here. A crucial element of my position will be the conclusion that, according to the Tractarian conception of language, the meaning of propositions is not only independent of empirical subjects, but also the condition of their possibility. This amounts to a resolute adaptation of Frege's principled anti-psychologism on Wittgenstein's part.

The Tractatus' inner layer of language
Bertrand Russell, in his Preface, expressed the view that one of the main goals of Wittgenstein in the Tractatus was to elaborate "a logically perfect language" (TLP, p. x 3 ). This was mistaken. Interpreters have long ago shown that Russell projected the aims of his own philosophy onto his former student's work. The latter never sought to create a new artificial language. On the other hand, however, the assumption of the absolute strictness of logic inherited from Russell and Frege spurred Wittgenstein to move towards a picture of language according to which the outer layer (common talk) disguises its inner logical core (thoughts). From this point of view our language -any language we can speak -is generally a two-layered phenomenon. Here I would like to point out that the Tractarian inner layer plays a similar role 3 All quotations and references to the Tractatus are indicated using the abbreviation "TLP", accompanied by the number of a thesis or a page (according to Wittgenstein, 2001). to that which so-called "primary language" (primäre Sprache) played in the short but distinctive phase of development of Wittgenstein's thinking that ran from February to October 1929. In other words, "primary language" amounted to a more comprehensive and explicit development of an idea already present in the Tractatus.
The idea of primäre Sprache is related to Wittgenstein's project (or perhaps we should say, outline of a project) of phenomenological research. Traces of such research appear in his 1929 notebooks, 4 as well as in the text Some Remarks on Logical Form (Wittgenstein, 1929) dating from the summer of that year. There are only a few of them, so this phenomenological phase in his thinking has not attracted much in the way of broader interest from scholars. According to Ray Monk's calculation, as of 2014 only 15 commentaries had been devoted to it (cf. Monk, 2014, pp. 335-336).
Wittgenstein's phenomenology was an attempt to examine the actual logical form of propositions expressing sensory phenomena, which he undertook in the wake of having conceded the points made by Ramsey in the context of the latter's criticism of logical atomism (cf. Zahavi, Overgaard, 2008, p. 63). The author of the Tractatus still held to his diagnosis that the grammar of common talk veils the real structure of thoughts, and he also continued to aim at the elaboration of a new, more perspicuous symbolism. However, the means for reconstructing the grammar of molecular propositions could no longer remain purely a priori in character. As he wrote in Some Remarks on Logical Form: The idea is to express in an appropriate symbolism what in ordinary language leads to endless misunderstandings. That is to say, where ordinary language disguises logical structure […], we must replace it by a symbolism which gives a clear picture of the logical structure […]. Now we can only substitute a clear symbolism for the unprecise [sic] one by inspecting the phenomena which we want to describe, thus trying to understand their logical multiplicity. That is to say, we can only arrive at a correct analysis by, what might be called, the logical investigation of the phenomena themselves, i.e., in a certain sense a posteriori, and not by conjecturing about a priori possibilities.
[…] An atomic form cannot be foreseen. And it would be surprising if the actual phenomena had nothing more to teach us about their structure. (Wittgenstein, 1929, p. 163) "Ordinary language" (gewöhnliche Sprache) is, in the course of the 1929 notebooks, also referred to as "physical language" (physikalische Sprache), "our language" (unsere Sprache) and "secondary language" (sekundäre Sprache). It is used to describe and discuss something Wittgenstein called the "second system" (zweite System) -meaning just our wellknown world of familiar physical objects. However, from the philosopher's point of view at that time, the latter objects were themselves theoretical postulates of some sort, as he reserved a less elusive mode of existence for just the phenomena that made up what he called the "first system" (erste System). 5 The author of the Tractatus focused on visual impressions, so his considerations usually concerned visual space. 6 According to him, this space (as well as the whole "first system") was adequately captured by the deeper layer of language -so-called "phenomenological language" (phänomenologische Sprache) or "primary language" (primäre Sprache). He held that any knowledge of the structure of phenomena themselves -including the structure of visual space -requires knowledge of the structure of the deeper layer of language, in that the former is reflected in the grammar of the latter. By examining the form of common talk alone, then, we cannot come closer to knowing the limits of meaningful description of reality, as this form lacks the required logical multiplicity 7 -it will only ever be either predicative or relational. Such limits can only be made explicit through a consideration of the form of "primary language".
The assumption of an intimate bond between the deeper layer and phenomenal reality led Wittgenstein to the conclusion that primäre Sprache could not be directly accessible to subjects. As he pointed out, what we know first and directly is our language, and our familiar physical space. 5 Strictly speaking, "ordinary language" also serves to describe the first system, but uses a hypothetical mode of presentation (MS-105, p. 108). 6 Wittgenstein's account of visual space during his phenomenological period will not be discussed here in detail. A reader interested in this topic may find useful e.g., the paper "Phenomenological Language and the Description of Visual Space" by Marcelo Carvalho (2013). 7 This important notion will be discussed in a subsequent part of this paper.
Phenomena themselves are not "ours": we do not own them. The philosopher stressed that "visual space essentially does not have an owner" . For this reason, both gaining access to it and recognizing the proper structure of phenomenological language require some exertion. 8 In consequence, one should admit that "primary language" does not belong to anyone, and that the justification for its existence is transcendental -this being a condition for the possibility of "physical language" (so that it is for this reason that the former counts as primary, the latter as secondary). 9 However, it is not clear what, for Wittgenstein during his phenomenological period, the actual method of examining the grammar of primäre Sprache was to be. After he had abandoned the idea of "primary language" at the end of 1929, he still tried to reach phenomenal structures through an examination of the syntactic invariants exhibited by the sense-data expressions present in natural language (cf. p. 205f.).
The thesis that the "primary language" of the phenomenological period is, in principle, a development of an idea already present in the Tractatus has been put forward, albeit only briefly, by Krzysztof Rotter (2006, p. 80). I believe that his intuition is to a certain extent correct; however, it requires further specification and also one fundamental qualification. The latter comes with the realization that the interpretation proposed by Merrill and Jaakko Hintikka thirty years ago (1986) and repeated later by Byong-Chul Park (1998), according to which the Tractatus was itself a work of phenomenology, cannot withstand criticism. 8 Ray Monk has pointed out a certain resemblance between this idea and the Husserlian phenomenological reduction, i.e., to the bracketing out of the "natural attitude" which, according to the founder of phenomenology, also requires us to perform a certain sort of mental act. As Monk suggests, by means of the metaphor of a picture on a screen and pictures on a film reel , Wittgenstein wanted to say that in the context of our normal, physical attitude, we speak about things in time and physical space, while in direct experience what we have are timeless phenomena located within visual space (Monk, 2014, p. 329). However, other commentators have argued that the resemblance between the Wittgensteinian and Husserlian conceptions is merely an apparent one (Vrahimis, 2014, p. 345). 9 As Jaakko Hintikka has rightly pointed out (2011, p. 167), the "first system" (phenomena) was also intended to serve as the basis for the "second system" (physical objects) from an ontological point of view: "The world we live in is the world of sense data", said Wittgenstein during one of his lectures in Cambridge in the thirties, "but the world we talk about is the world of physical objects" (Wittgenstein, 1980, p. 82).
The Hintikkas and Park viewed the early Wittgenstein as a phenomenologist because, so they argued, he was preoccupied with the problem of what it is that is directly given to us, and to what extent it is so. In order to adhere to this interpretation, one must accept that Tractarian simple objects are contents of experience. To be sure, as Park argues, the theory of reference presented in the Tractatus is a critical development of sorts of Russell's conception dating from the 1910s (cf. Park, 1998, p. 34). 10 But what speaks against this phenomenological interpretation of the early Wittgenstein is that the Tractatus offers a rather skimpy specification of simple objects (although in Notebooks 1914(although in Notebooks -1916 the philosopher did consider some more substantial answers, including the idea that they are minima sensibilia), and what it says about them -that they are colorless, indestructible, and have practically no properties -seems to explicitly preclude just this interpretation. (This is in all probability the reason why such an interpretation has proved rather unpopular.) As we shall see, when we come to address the second thesis of this paper, there are persuasive reasons for recognizing Tractarian objects as being something like Heinrich Hertz's mass-particles, i.e., as being something that is required to exist just by virtue of being entailed by the structure of "primary language" itself. 11 The Tractatus points clearly to the idea of there being a surface layer and a deeper layer where language is concerned. It uses the metaphor of "clothes": our expressions are "clothes" covering a "body" corresponding to the proper structure of our thoughts. This structure is the inner core of every possible language, but it is also the very condition for the possibility of its having sense. At the same time, it is not directly accessible to subjects; they themselves have no idea "what each word means" (TLP 4.002).
Beneath the layer of common talk in language, and the language of scientific hypotheses -actually, we should talk here about languages in the 10 It is also worth noting another phenomenological interpretation, which is presented by Mathieu Marion in his paper "Wittgenstein and Brouwer". Marion assumes that for the early Wittgenstein it is not things but propositions that are directly given to us (cf. Marion, 2003, p. 110). Criticizing this view would require a different line of argumentation from the one directed here against the Hintikkas and Park.
11 Strictly speaking, Hertz posits no requirement for the existence of mass-particles; he puts forward a certain picture of mechanics while not excluding the possibility of another depiction being more adequate (cf. Hertz 1894, pp. 48-49). Here lies the difference between his Prinzipien and the Tractatus (this insight I owe to Joshua Eisenthal). plural, in that there are different national languages -there is a universal logical structure, the structure of thoughts. The Tractatus explicates this fundamental structure as a series of forms that are truth-functions of elementary propositions. The definition of this series, which is at the same time the general form of the proposition, is given by thesis 6: "[ ͞ p, ͞ ξ, N( ͞ ξ)]". 12 One of the ramifications of this is that each meaningful proposition is, in its deeper layer, a truth-function of elementary propositions (cf. TLP 5). So the latter are the "bricks" of Tractarian primary language, being composed of names that simple objects correspond to. It may be shown that such names (but also the objects, as I will argue in due course) are strictly abstract postulates, and that it makes no sense even just to ask about what they look or sound like. Note that just as with phenomenological primäre Sprache, Tractarian "primary language" was supposed to be capable of expressing reality without any ambiguities or confusions. "Secondary language", on the other hand, was to lack clarity 13 . Nevertheless, the meaning of our sentences depends on their unequivocal analyzability -a process that should end with the appearance of elementary propositions. As thesis 4.52 says: Propositions comprise all that follows from the totality of all elementary propositions (and, of course, from its being the totality of them all). (Thus, in a certain sense, it could be said that all propositions were generalizations of elementary propositions.) 14 That "certain sense" points to a problem, which is that Wittgenstein gives no example to illustrate his analysis, i.e., no instance of a transition from our sentences containing familiar parts of speech and grammatical structures to elementary propositions consisting of the names of simple 12 In this definition " ͞ p" means all elementary propositions, " ͞ ξ" means a general member of a series, and "N( ͞ ξ)" means a general nexus member being a result of the N-operation applied to a previous member.
13 However, the Tractarian secondary language, unlike the phenomenological physikalische Sprache, would not lack precision. Colloquial sentences, if are to express their sense, must not differ from their fully analyzed counterparts in regard to logical multiplicity. Wittgenstein modified his stance on this issue some time before 1929 as he became aware of non-logical necessary connections between elementary propositions (I noticed this thanks to Joshua Eisenthal).
14 As is well known, universal generalization is, according to the Tractatus, a kind of truth-function: namely, multi-part conjunction.
objects. Neither are we given any real example of the structure of an elementary proposition. 15 Therefore, some interpreters suggest that the Tractarian talk of analysis, in its entirety, serves merely to explicate the thesis that our sentences have sense. They are thus inclined to view the whole idea of Tractarian "primary language" as being illusory. Such interpreters also tend towards the thesis that elementary propositions play no role except in the context of the analysis of our common talk, holding that their only role is that of pointing to essential features of standard sentences (cf. Eisenthal,unpublished,p. 3; see also Kremer, 1997).
Undoubtedly, Wittgenstein should not have sought to develop a theory of elementary propositions, if he had wished to remain faithful to his own meta-philosophy. The point is that if one gives autonomy, or even priority, to such a thing as the inner layer of language, one is abusing the scope of what philosophers are entitled to pursue as set out in theses 4.111-4.116: one is not engaged in the "logical clarification of thoughts", but rather in some sort of transcendental speculation. The Tractarian view of language requires such speculation, because it assumes the transcendental nature of Russellian-Fregean logic (which is affirmed explicitly in the thesis 6.13). This assumption gives rise to a need to elevate the status of logical relations above all others. The latter may be grounded within our practice of using language (in common talk or in science), but the former cannot be, so one is forced to seek a firmer foundation. (As will be discussed in the third part of this paper, one is also forced to talk about the limit of the world and the metaphysical "I".) The same assumption leads to an acknowledgment of the Russellian model of analysis. The "merit" of the British philosopher, admitted by Wittgenstein directly in the thesis 4.0031, is to have shown in "On Denoting" that the grammatical form of a (common) sentence may be merely apparent, so we need to work to seek out the "real" logical form of our utterances (cf. Russell, 1905). The doctrine of the Tractatus holds that if the latter exists at all (i.e., if a common sentence possesses sense), then it must be a truth-function of elementary propositions.
According to the early Wittgenstein, the inner layer -the Tractarian primäre Sprache -is thus an inherent part of any language. It will be universal, and common to all languages that can express any sense whatsoever. Due to its particular features, this layer is not given to us directly, and perhaps we are unable to reach it at all. Nevertheless, it does exist. It shows itself to some extent through the logical regularities of surface expressions, where how far it does so will depend on the particular logical apparatus employed in unsere Sprache. Because we can bring order to the surface layer of language through the use of proper notation, the underlying logical relations can be rendered more perspicuous. Wittgenstein thought that the greatest achievement of his German and British predecessors was to have created notations that give a certain insight into the real structure of the symbolism, and he understood his own task as being that of following this up with further advances in the same direction. 16 That was what his early understanding of the idea of perspicuous representation looked like.

A Hertzian interpretation of the Tractatus
The nature of Tractarian simple objects has long been a debated issue. Among others things, it has been disputed whether these can only be of a material sort or can be such abstract entities as predicates and relations, whether the sum of all objects makes the objective -and thus subject-independentsubstance of the world or is, instead, a kind of transcendental form projected onto reality by the knowing subject, and whether they can be identified with elementary particles or -as a couple of the authors mentioned in the preceding section of this article have claimed -are rather phenomena. There is also a standpoint according to which all the above questions lack sense, as the Tractatus offers no ontology -nor an epistemology of simple objects.
Among both proponents and opponents of an ontological reading of the early Wittgenstein's philosophy, there are interpreters who try to solve the simple objects conundrum by examining similarities between the Tractatus and Die Prinzipien der Mechanik, by Heinrich Hertz. Wittgenstein himself explicitly pointed to Hertz as a significant source of inspiration: he had acquired a high level of knowledge and appreciation of the Prinzipien even before arriving in Cambridge to study philosophy with Russell (cf. Monk, 1990, p. 23). Moreover, Hertz's name appears twice in the text of the Tractatus.
Gerd Graßhoff, in his paper entitled Hertzian Objects in Wittgenstein's Tractatus (1997), defended an ontologically-oriented materialist interpretation of the early Wittgenstein's philosophy, arguing that it follows from the assumption of Hertz's influence. A similar idea was proposed earlier by James Griffin (1964). Both authors suggested that simple objects should be understood as counterparts of Hertzian material points. This article is not the place for a broader critical discussion of these proposals, but it should be highlighted that as their starting point they assume an ontological interpretation of the Prinzipien der Mechanik. However, as has been argued by some other authors, to ascribe ontological intentions to Hertz is at least as disputable as with the early Wittgenstein.
Joshua Eisenthal and Sara Bizarro have each argued independently for the so-called logically-oriented Hertzian interpretation, according to which Tractarian ontology is nothing but an interpretative artifact. As Bizarro shows, the same arguments that can be used to refute the idea that simple objects are a kind of sense-data (briefly presented in the previous part) speak against Griffin and Graßhoff (cf. Bizarro, 2011, p. 155f.). Both she and Eisenthal attempt to make explicit the fact that if one is going to look for Hertzian counterparts of Tractarian objects, one should turn to the notion of mass-particles (Massenteilchen).
How should this Hertzian notion be understood? Eisenthal (unpublished, pp. 8-13) presents a detailed elucidation of it, deconstructing along the way the supposed ontology of the Prinzipien. As he points out, the main goal of the work was to give the general form of every possible description of dynamic systems, in terms free from metaphysical confusions of the kind generated by, for example, questions about the "real nature" of physical force. So, it can be reasonably asserted that Hertz meant to present rules for the construction of models of these systems: rules that would allow one to achieve conceptual clarity within mechanics. According to him, the essential features of any given system were the number and type of its degrees of freedom. Those would be reflected in the configurational space ascribed to it, where one and the same configurational space could be related to infinitely many different physical systems, such that each of them could be considered a model of the others.
Despite the fact that almost all dynamic systems are subject to some external forces, Hertz postulated that we should treat all systems as free; that is, as independent of external influences. (His goal was to demystify the notion of force in physics.) What classical mechanics describes in terms of forces external to a given system, the Prinzipien der Mechanik represents by so-called "hidden masses". The system in question was to be regarded as partial, i.e., as forming part of a larger free system, whose other remaining part (represented by a hidden mass) was itself invisible and unanalyzable while serving to explain the movements of the "visible" part. As Eisenthal stresses, symbolic pictures of dynamic systems constructed within such a Hertzian framework do not make any claim to ontological adequacy.
[T]he sole requirement on a picture is that its consequences represent the consequences of what it pictures. What Hertz strenuously emphasizes is that, on his view, the representative content of a theory does not go any further than this: 'we do not know, and we have no way to learn, whether our conception of things conforms with them in any other way, except in this one fundamental respect alone'. (Eisenthal,unpublished,p. 9) So, neither material points nor Massenteilchen were conceived as particular physical elements of any sort: instead, they were just conceptual tools that allowed Hertz to create mathematical models of dynamic systems, such as would fulfill the requirement of predictive efficacy. The Prinzipien der Mechanik defined mass-particles as properties of spatio-temporal points, in that they served as a measure of mass ascribed to a given space. Again, the job of geometrical mass systems created according to Hertzian postulates was not to reflect actual physical structures, but rather to enable one to foresee the behaviour of fragments of the physical world considered as dynamical systems. Thus, it does not make sense to ask the question posed by Graßhoff (1997, p. 105) -what does it really mean that certain spatio-temporal places have certain attributes (Massenteilchen)? The mass-particles were just the most elementary components of the models described within the framework of the Prinzipien der Mechanik.
There is one place in the Tractatus where Wittgenstein directly recalls the title of Hertz's book. It occurs in thesis 4.04, which says: There must be just as much that is distinguishable in a sentence as there is in the situation that it represents. The two must possess the same logical (mathematical) multiplicity. (Compare Hertz's Mechanics on dynamical models.) As is rightly pointed out by Eisenthal, the thesis suggests that logical (or mathematical) multiplicity is, in the context of the Tractarian conception of logical picturing, a feature analogous to the multiplicity of dynamical models in the Prinzipien der Mechanik. The equivalence of Hertzian systems consists in their possessing the same multiplicity and type in respect of their degrees of freedom (i.e., in the identity of their configurational spaces), while that of Wittgensteinian propositions consists in the identity of the corresponding sets of all their logical relations to other propositions (Eisenthal,unpublished,. This is what is meant by thesis 3.4, which says that "the proposition determines a place in logical space". 17 As regards molecular propositions, inter-propositional relations result from their sharing the same atomic building blocks: the corresponding truth-functions have the same elementary propositions as their arguments. So the truth-value of a molecular proposition can be dependent on the truth or falsity of other molecular propositions. By contrast, at the level of elementary propositions there are no such relations. However, given the fact that various elementary propositions contain the same names, relations between senses of these propositions will follow. As is pointed out by Marie McGinn, what constitutes the meaning of a name is its occurrence in a series of different elementary propositions which may be true or false in particular situations (McGinn, 2006, pp. 88, 194f.).
The proponents of anti-ontological interpretations point out that the relation between a Tractarian name and a simple object is distorted by a certain confusion, as Wittgenstein gave in to the temptation to identify the meaning of simple signs with their bearers; namely, with objects (McGinn, 2006, p. 114f.). Indeed, he noticed this confusion himself and criticized it in several paragraphs of the Philosophical Investigations (Wittgenstein, 2009, § §39-41). What followed from this confusion was precisely his striking account of objects -i.e., of their non-complexity and indestructibility. This was because the postulate of the determinacy of sense assumed by the author of the Tractatus (TLP 3.23) required that the meanings of simple signs be unquestionable as regards their durability.
On the other hand, if we embrace the suggestion put forward by Bizarro and Eisenthal then we arrive at an interpretation according to which the scale of this confusion is significantly smaller: Tractarian simple objects are by no means independent elements of reality, but only certain aspects of reality determined by the grammar of elementary propositions. 18 To be more precise, they are the simplest elements that we may distinguish within reality understood as a correlate of language, and so perform exactly the same role as the Massenteilchen do within the Hertzian system.
It is worth noting that if Wittgenstein had trusted completely in his intuitions prompted by the Prinzipien der Mechanik, he might well have come up with a conception of meaning focused on use as early on as in the Tractatus. 19 He would not have had to give any elucidations regarding the nature of simple objects: the meanings of the smallest distinguishable parts of logical pictures would have been determined by nothing more than the mutual relations obtaining between the pictures themselves. The reason Wittgenstein chose another way was his respect for Frege and Russell -and, in particular, for the latter's "merit" in having formulated the paradigm of analysis put forward in On Denoting, as was already mentioned above. The point was 18 As Eisenthal writes: "On a logically-oriented interpretation, Tractarian analysis uncovers whatever forms of elementary sentences and forms of names are needed in order to capture the manifest logical relationships among colloquial sentences. On this view, elementary sentences and the names of simple objects do not have significance apart from the analysis of colloquial sentences" (Eisenthal,unpublished,p. 18). 19 Nevertheless, this work does carry the seed of a meaning-as-use conception, as has been pointed out by Michael Kremer (1997), Cora Diamond (2000), James Conant (2000) and others. The first to notice this was Hide Ishiguro in 1969, when she stated that the Tractarian conception of meaning is in fact very similar to the meaning-as-use conception associated with the later period of Wittgenstein's activity, with the difference being "that the Tractatus concept of 'use' is much less comprehensive than in the Investigations" (Ishiguro, 1969, p. 21). that if it was reasonable to look for the "real" form of our utterances, then there had to be a criterion for that form's counting as "real". Such a criterion need not be a definite and subject-independent form of reality, as ontological interpretations of the Tractatus propose. One could, instead, simply proceed on the assumption that there is a universal and fully logically adequate deep structure of language (and thought) hidden beneath the surface of unsere Sprache. That structure will itself ensure super-rigid references to its most elementary parts (i.e., names in atomic propositions), imparting a form to reality that is such as to ensure that the latter breaks down into atomic facts construed as combinations of simple objects.

Radical anti-mentalism, or subjectless transcendentalism
Many important interpreters of the early Wittgenstein take him to have been a mentalist. According to them, he thought that the sense of a proposition of language is inherently correlated with the fact that a subject has thought that proposition. Such a reading is proposed, among others, by Norman Malcolm (1977, pp. 120-164), and Peter Hacker (1999). Its point of departure is thesis 3.11, which says that we use signs of propositions as projections of possible situations, and "[t]he method of projection is the thinking of the sense of the proposition". The immediately ensuing thesis (3.12) then formulates the claim that propositional signs express thoughts.
However, a mentalistic reading of these excerpts from the Tractatus has proved by no means uncontroversial. It has prompted objections from Rush Rhees (1970, p. 39), Peter Winch (1995 and Cora Diamond (2013). It is beyond the scope of the present paper to discuss these views in detail. I shall instead point to just one of the serious obstacles facing such an interpretation: if it were true, the Tractarian critique of "modern epistemology" presented in theses 5.541-5.5423 would then look highly doubtful. 20 One version of that kind of approach to epistemology had been proposed by Russell in 1912Russell in -1913, in the form of his multiple-relational theory of judgment. He had assumed that the subject of judgment is external 20 An alternative is to assume that the theses 3.11 and 3.12 are framed in language of a higher-order level than that which the rest of the Tractatus talks about. However, this would spoil Wittgenstein's solution to the problem of set-theoretical antinomies.
to language, and functions as a binding agent, linking words together into sentences. Wittgenstein had refuted this solution as early as summer 1913 (cf. Wittgenstein, 2008, p. 40). In his wartime notebooks, moreover, he had argued that Russell's theory does not furnish us with criteria for telling apart sentences that are properly formed on the one hand, and such ill-formed pseudo-sentences as "the table penholders the book" on the other (cf. Potter, 2008, p. 121f.). Later, in the Tractatus, he presented a view to the effect that such nonsensical mixtures of words cannot be thought. One aspect of this view is the thesis that so-called "intensional" contexts may, in the course of analysis, be replaced by propositions that lack such contexts, without any change to the sense of the utterance in question. As he wrote in thesis 5.542: But it is clear that "A believes that p", "A thinks p", "A says p", are of the form "?p? says p": and here we have no co-ordination of a fact and an object, but a co-ordination of facts by means of a co-ordination of their objects. 21 It follows from thesis 5.542 that thinking subjects are not external to language, as had been assumed by Russell. For, given that "A thinks p" has, in fact, the form "'p' thinks p", then if 'p' is a fact which symbolizes p and which is related to it by means of projection, as thesis 3.11 claims, then A should be understood as a set of propositions (beliefs) requiring language as a condition of its own possibility. 22 However, this means that the thinking subject that thesis 5.542 talks about -the common human being to whom we ascribe certain beliefs, including the human being qua referent of the pronoun "I" when that pronoun is construed psychologically -cannot be the subject of an act of thinking the sense of propositions, i.e., some act such as would yield a method of projection responsible for imparting meaning to signs.
This conclusion brings us to the transcendental interpretation of the Tractatus. According to its classical version, there is a supra-empirical subject linked to various mentions of the "I" construed in non-psychological terms, and a metaphysical subject, this latter being the limit of the world, as in thesis 5.641. Such a subject was, in a mysterious way, to be identical to both the narrator of the Tractatus and the world itself (theses 5.6 and 5.63), and so would constitute the inexpressible truth of solipsism. 23 Yet, if we take seriously Wittgenstein's remarks on philosophy in theses 4.111-4.116, this classical story of a supra-empirical subject loses its cogency. For the proper goal of the Tractatus is not then to construe theories based on a priori insights into the nature of reality, 24 but rather to elucidate certain problematic notions. One of these is the notion of "I" construed in a non-psychological sense, or the metaphysical subject. Thesis 5.641, read as an elucidation, says that there is, in fact, a certain context in which we are inclined to use such notions. This context is, however, very peculiar: it is a philosophical discourse about the world as a whole. In such a framework, the metaphysical subject is the limit of the world.
It may be said that the Tractarian notion of the limit of the world (which is also the limit of meaningful language and thinking -see 5.6 and 5.61) is limited itself, for there is no place for such a notion in a perfectly logically ordered language. This is because it is neither the name of an object or possible complex of objects, nor a formal concept (a concept expressing the character of a certain variable or the general term of a certain series of forms). Its dubious status is revealed in the Preface, where Wittgenstein writes about drawing a limit to thinking and at the same time excludes any associations with the idea of dividing up some uniform area (cf. Stern, 1995, p. 76f.). It is clear that the author of the Tractatus uses the concept of limit 23 Many interpreters take this approach to reading the early Wittgenstein, e.g., Hans Sluga (1996) and Richard Brockhaus (1991). Pasquale Frascolla goes a few steps further, and identifies the metaphysical subject with God (Frascolla, 1994, p. 30f.). 24 It is worth noting that the Wittgensteinian critique of transcendentalism, which was clearly expressed by him both in the 1930s and in later remarks, did not focus solely on epistemic issues. Wittgenstein considered transcendentalism in its many forms to be a manifestation of the decay of European culture (cf. Markewitz, 2019). only provisionally, to express his aims, and is happy for it to be thrown away afterwards (cf. TLP 6.54).
The real reason the young Wittgenstein required the concept of limit was for his overall conception of logic, which is transcendental in spirit. It is true that the Tractatus was an important step forward on the way to removing confusions about the character of logical relations, compared to the metaphysical conceptions of Frege and Russell. But it did not go all the way. The early Wittgenstein claimed, on the one hand, that the theses of logic do not express anything, and that logic cannot be described as it sets the norms for any description itself. On the other hand, though, he also claimed that logical norms are unshakable and absolutely strict. 25 The actual empirical reality of our lives cannot itself justify any such absolute strictness, hence the need for such Tractarian metaphors as "the scaffolding of the world", "the essential in a symbol", and "the essence of the notation", expressing a belief in a deep layer of symbolism (as discussed in the first part of this paper). Only after his rejection of the postulate of "the crystalline purity of logic" (Wittgenstein, 2009, §107) was he in a position to remove this metaphysical residuum, but the philosopher did not accomplish this before the thirties.
Returning to the Tractarian concept of limit, we may also note that in the first instance it serves to distinguish sense from nonsense: in order to eliminate misunderstandings that persist within unsere Sprache, we need to be able to separate them from what is truly meaningful. We do not accomplish this by drawing some line of demarcation, but rather by elucidating the structure of the inner layer of language (the Tractatus' primäre Sprache, as I have called it, following the example of Rotter). The aim of the latter is to make the relationship between the two layers transparent through the introduction of perspicuous notation into unsere Sprache -with this sufficing to fully disclose our nonsensical utterances. Once this goal has been fulfilled, there will be no further use for the notion of limit.
All this means that the non-psychological notion of "I" is provisional too. It has its place within philosophy understood as the activity of logically clarifying thoughts (see TLP 4.112), and serves that purpose, 26 but certainly only if not treated as a subject of description. 27 For if the transcendental subject is to be understood as the limit of the world, and that limit is not external but internal, in that it is identical to the logical structure of primäre Sprache, then one may say that the non-psychological "I" is ultimately a hypostatization of a certain aspect of the activity of the inner layer of language. No wonder that Wittgenstein came to the conclusion that there is a truth in solipsism, but that it cannot be expressed! For its expression would have had to have been as follows: "primary language" is itself its own metaphysical subject, for it thinks the senses of elementary propositions and thus provides the method of their projection, and since logic is a unity, and in consequence primäre Sprache is universal, so there can only be one non-psychological subject. 28 Yet such an expression would have been nonsensical: the early Wittgenstein plainly does not construe the symbolism as constituting a subject of sorts in and of itself, as it has no will, no needs, and no goals to achieve. 29 The only thing that inclines us to construe the symbolism in such terms is its auto-projective function.

Conclusion
The interpretation offered in the present paper may prove controversial, not only because it is based on interpretative judgements that are open to 26 It may be noted that another provisional notion of the kind used in the Tractatus is that of the world as a whole, employed in theses 5.6, 6.41, 6.43, 6.45 and commentaries upon these. 27 Note that Tractarian elucidations pass from non-psychological uses of "I" to possessive pronouns: "The I occurs in philosophy through the fact that the »world is my world«" (TLP 5.641).
28 Cora Diamond has pointed out that Wittgenstein uses the impersonal mode of speech when he writes about thinking and the method of projection. She argues that he does so deliberately in order to dissuade readers from adopting a mentalistic interpretation (cf. Diamond, 2013, p. 154). 29 Some interpreters have presented even bolder theories. Jaakko Hintikka, in one of his 1958 papers, proposed the thesis that the Wittgensteinian metaphysical subject is just the sum total of language (Hintikka, 1958, p. 89). However, Hintikka also identified the metaphysical subject with a factual subject, such as the subject A from thesis 5.542 (Hintikka, 1958, p. 90). dispute, but also because it holds that the Tractatus Logico-Philosophicus is incoherent in that it develops intuitions that ultimately cannot be made to fit together. It seems reasonable to assert that if this last point holds, then there is no definite answer to the question of the correct interpretation of Wittgenstein's early philosophy. The only thing which can still then be said is that a certain multiplicity of coherent but mutually exclusive interpretations is maximally convergent with what the philosopher wanted to say in the Tractatus.
Perhaps this is so. Nevertheless, the view presented here may seem quite reasonable. Its starting point is the conception of philosophy given in theses 4.111-4.116. Yet this conception itself faces an impassable obstacle: a requirement of supra-empirical durability on the part of the fundamental structures of language, where this is taken to follow from the very idea of logic. Therefore the Tractatus, contrary to its own declarations, must after all contain a certain minimal metaphysics postulating a transcendental structure. This structure will be the perfectly logically ordered inner layer of language. On the basis of functional resemblances to the particular conception of a phenomenological language adopted by Wittgenstein in 1929, it can be referred to as "Tractarian primäre Sprache".
The non-classical version of a transcendental reading of the Tractatus presented here reduces its metaphysical content to an absolute minimum, in that it rules out any "full-blooded" transcendental subject. Some classical transcendental readings treat the latter as one of the two poles of the intentional relation that constitutes symbols (meaningful signs), the other one being extra-linguistic reality. Instead, the non-classical version imports this intentional relation, so to speak, into language itself: more specifically, into its transcendental part -primäre Sprache. The only metaphysical part of this image is the part that the early Wittgenstein assumed to be metaphysical from the very beginning. "Primary language" "absorbs" the non-empirical subject -or rather, takes over the function of this subject -much as it "absorbs" the ontological structure of reality, which latter turns out to be secondary to its own grammar.
It should be stressed that just as in the phenomenological period, so also at the time of the Tractatus, "primary language" functions above the level of empirical subjects. They are responsible only for the "clothing", i.e., the surface layer of the symbolisms they use, whose complicated texture results from tacit social agreements. This layer can only have sense, can only reach