Core Language of Thought

Kyo Takano
Independent Researcher
Japan
[email protected]

Abstract Mentalese’s generational or interpersonal variabilities. Ac-

cordingly, in this article, I present a novel perspective on the
Fodor’s “Language of Thought” (also called Mentalese) Language of Thought hypothesis, incorporating such vari-
hypothesis—that we use a language-like, syntactic system abilities caused by several factors. Specifically, I argue that
in processing thoughts instead of natural languages—caused individuals have their own “Idiolects of Thought,” which
controversies in philosophy, psychology, and linguistics.
While many researchers proposed their perspectives, both for
partly vary from individual to individual and share a large
and against, none of them addressed its evolutionary and so- portion of common features that I would call the “Core Lan-
ciocultural variabilities as far as I know. Accordingly, I pro- guage of Thought.” This contrast is equivalent to the fact
vide a novel perspective on the hypothesis by discussing the that, for example, all English speakers do not use exactly the
interplay among our cognition, natural languages, evolution, same English but their idiolects—English with slight and
and sociocultural factors. First, like one speaks personalized unique variations. In providing my argument, first, I will
English as his/her idiolect, individuals have their own “Id- present premises for my argument from evolutionary, socio-
iolects of Thought” as a derivative of Mentalese. Second, cultural, and cognitive perspectives. Subsequently, I present
while Mentalese varies from individual to individual, “Core how humans can have variational Mentalese, a language-like
Language of Thought” is cognitively fundamental and largely system to process mental representations of concepts, both
shared across individuals.
evolutionarily and socioculturally with illustrations.
While people often think they use English or other natural
languages, it is not a plausible explanation of our thought
Premises
processes. A language is itself is a mere medium to repre- Connectionism
sent thoughts comprised of their constituent concepts, and In discussing the Language of Thought hypothesis, as a
our brains enable encoding/decoding of neural representa- premise, I take a connectionist approach and regard that con-
tions corresponding to such. Instead, like infants without cepts in thoughts are distributed representations in the brain.
words can make simple inferences, it is more plausible to Here, I interpret Mentalese as a metaphorical projection of
believe we possess a mind-internal system—probably in- how concepts are neurally and thus high-dimensionally rep-
nate—to process thoughts. resented and processed; in other words, Mentalese is funda-
Fodor (Fodor 1975; 2008) named such a system the mentally rooted in neural representations of concepts, and
“Language of Thought” (also Mentalese). The Language of their relations can still assume it in the form of language.
Thought hypothesis states that thoughts are held and pro- Although Fodor and some scholars disagree with the con-
cessed in Mentalese, a system that has a language-like syn- nectionist view, particularly because it seems to undermine
tax and compositional semantics. Like natural languages, a the systematicity of Mentalese as an information-processing
Mentalese sentence in Fodor’s definition supposedly con- system (Fodor and McLaughlin 1990; Fodor, Pylyshyn, and
sists of token representations of concepts, and its meaning is others 1988), I refute this objection for the following rea-
systematically determined by those representations and the sons. First, Mentalese is just a language-like syntactic sys-
way they are composed together. For example, an English tem but an actual language. While natural language expres-
phrase YOUR DOG RAN AWAY can denote that a specific sions can be analyzed and explained, that is only because
dog in your house ran away to somewhere, as the meaning our linguistic system can decode them into Mentalese ex-
can be reconstructed to its mentalese expressions through pressions, and thus Mentalese itself need not necessarily
corresponding mental representations of YOUR, DOG, RAN have this property of natural language. Second, such dis-
AWAY , and the subject-verb structure. tributed representations do not undermine the systematicity
Although many scholars—philosophers, psychologists, of thoughts, whereas enjoying flexibility and productivity.
and linguists—debated over this controversial hypothesis, to While an explainable system would feel intuitively system-
my knowledge, there has been no argument concerned with atic, Mentalese neither equals natural languages nor has to
be as highly explainable as English. Simultaneously, sys- Core Language
tematicity can be achieved without high explainability when ･･･
of Thought
nodes of mental representations can constitute a thought
state in a structure-sensitive manner (Chalmers 1990; 1992;
Niklasson and Van Gelder 1994; Markman and Dietrich Natural languages ･･･
2000). Finally, connectionism also explains various men-
tal representations/processes, including concepts that words G
can denote and logical structure, perception, mental map,
etc. (Rescorla 2019). For more details of the controversy, Individual sample
see the review by Rescorla (2019).
Idiolect
Brain-Language Interaction ･･･ ･･･
of Thought
Second, our cognition and natural languages are strongly de-
pendent on each other; the usage of natural languages influ-
ence the way we think (language relativity) (Kay and Kemp- Idiolect ･･･ ･･･
ton 1984; Hussein 2012; Boroditsky 2001; Alderson-Day
and Fernyhough 2015), and the way we think shapes our
use of natural languages. One explanation for such a mu- T
0 (at birth)
tual dependence is that, over generations, languages evolved
to convey more and more detailed social messages as a Figure 1: Evolutionary and Translanguaging variations of
brain-friendly communication tool, which required humans Language of Thought, with a constant environment as-
to have language-friendly cognitive functions (Evans and sumed. The upper half depicts the slight differentiations of
Levinson 2009). both the Core Language of Thought and natural languages
influencing each other over generations. The bottom half
Translanguaging shows how an individual shapes his/her Idiolect of Thought
Finally, I also assume that humans have innate cognitive and natural language(s) together. With innate neural regis-
capacities to learn human-level languages and that learn- ters for certain primary concepts, s/he learns to use natu-
ing/acquiring a language involves lots of factors including ral languages as his/her idiolect, which mutually influences
but not limited to linguistic, cognitive, and semiotic ones, his/her idiolect of thought.
as Wei (2018) argues. This is the concept of “Translanguag-
ing” (Wei 2018), which allows language users to constitute
their own idiolects by integrating different dimensions asso-
which I consider is the basis of Mentalese and also is largely
ciated with their social lives, including native language, ide-
innate. Like every English speaker uses personalized En-
ology, personal experiences, etc., beyond the linguistic do-
glish slightly deviated from other English speakers, I sug-
main. Put another way, for example, one’s native language
gest that Mentalese be considered in the same way; in other
is a mere determinant of his/her idiolect, among many other
words, Mentalese does not have to be totally consistent re-
factors. Since the concept well explains the sociocultural
gardless of thinkers as mentioned above. At the same time,
variability of natural language and mentalese, I will em-
while our Mentalese varies from individual to individual, a
ploy the Translanguaging perspective in expanding my ar-
large part of it is plausibly shared across individuals. This is
gument. Here, although Wei (2018) opposed Fodor’s Lan-
because we share fundamentally the same cognitive features
guage of Thought argument (1975), I do not fully accept the
shaped by evolutions. Borrowing Wei (2018)’s perspective
rejection because it was largely based on the assumption that
of Translanguaging, the variability of Mentalese can be con-
Mentalese exists independently of idiolects; in my argument,
sidered largely caused by a number of sociocultural factors,
I allow Mentalese to have variability associated with idi-
such as languages used in one’s everyday life, ideology, per-
olects. Translanguaging is also highly compatible with the
sonal experiences, cognitive capacities, etc., like Wei (2018)
above two premises. First, the connectionist view assumes
addressed how individuals learn to have idiolects as a re-
distributed representations of thoughts and thus language
sult of Translanguaging. Figure 1 metaphorically illustrates
use; this suggests that the use and acquisition of languages
evolutionary transformation and intra-generational variabil-
can plausibly extend to other interrelated cognitive domains.
ity of Mentalese. Core Language of Thought, as the repre-
Second, Translanguaging explains intra-generational inter-
sentative component of Mentalese, and natural languages are
action between brain and language when there is little liter-
fairly stable over generations; however, they still have slight
ature on the evolutionary interaction.
effects on each other in response to natural languages’ so-
cial utility, brain-language fit, etc. In contrast, one individ-
Core Language of Thought ual’s Idiolect and Idiolect of Thought grows rapidly; at birth,
Having provided these premises, I propose that we have even with no utterance ability, s/he expectedly has his/her
variational languages of thought—which I call “Idiolects own immature Idiolect of Thought as a derivative of Core
of Thought” —and that a significant proportion of it is Language of Thought, which is associated with the innate
cross-personally consistent as “Core Language of Thought,” capacity to learn natural languages. Over the lifetime, the
 English idiolect Japanese idiolect Overlay

I YOU 俺 おまえ

Word-level
TALK SPEAK 話す

Language-level

Figure 2: Pseudo feature embeddings of different Idiolects of Thought into 1D space on both word and language levels. Each
distribution curve represents a concept or a natural language with contextual variances in one’s Idiolect of Thought, and they
share common and context-dependent features over the overlapping areas. For the word-level embeddings, the top row shows
fairly distinct concepts of single- and second-person nouns, which are probably innate; they overlap only slightly within each
individual’s idiolect of thought, while there are also little differences between individuals. On the other hand, the second row
indicates that even if there are some deviations between concepts, they also share certain features. As a paradigm of such
perspectives, it could be inferred that languages also share certain features independent of individuals who speak different
idiolects and have different Idiolects of Thought.

individual learns to speak natural languages in the form of reduced to their lower-level constituents.
idiolect while also developing Idiolect of Thought in the mu- As a limitation, this argument can be true if and only if the
tual relationship. variabilities of Mentalese are accepted, and also our cogni-
tive characteristics are, in fact, fairly the same from individ-
To further address how Idiolects of Thought vary across ual to individual. Although I have no means to prove these
individuals, I metaphorically depict translanguaging vari- premises, based on my introspection, they are intuitive and
ability of neural representations in Figure 2. Although this make the Language of Thought hypothesis more plausible
illustration lacks authenticity as words and languages cannot as a theory of our thought processes. Because the purpose of
be embedded into such a low dimension realistically, it visu- this paper is to present the Core Language of Thought and
alizes qualitative similarities/differences of their neural rep- Idiolects of Thought, I did not fully address past controver-
resentations both intra- and inter-personally. In the top ex- sies on Metalaese.
ample, concepts like I and YOU are very distinct as they are
plausibly determined by our evolutionary need to distinguish Conclusion
the most important communication entities. This example is
certainly cross-culturally consistent like the Japanese lan- Discussing the nature of natural languages and Mentalese
guage has its equivalents. Accordingly, when overlaid, there from evolutionary, sociocultural, and cognitive perspectives,
are only small interpersonal gaps for neural representations in this paper, I proposed that every individual have his/her
of these concepts. On the other hand, TALK, SPEAK, and own Idiolect of Thought as a derivative of Mentalese while
their Japanese equivalent have similar but slightly different sharing the Core Language of Thought with others as a fun-
connotations. Regarding these gaps, I would explain that the damental component of Mentalese. As far as I know, this
overlapping distribution of features represents constituent paper is the first to assess Mentalese—also called the Lan-
concepts shared by these words’ neural representations and guage of Thought hypothesis—on its evolutionary and so-
that the differences are a product of slight differences of ciocultural variabilities together with those of natural lan-
translanguaging experiences. Finally, summing up these ob- guages. The variability of Mentalese might have implica-
servations, I contend that different individuals—either from tions for several fields such as linguistics, cognitive psychol-
the same or different cultures—fundamentally consist of ogy, Natural Language Processing (a branch of artificial in-
largely the same features. Even though some neural rep- telligence), language education, etc.
resentations for equivalent concepts might seem somewhat
different like in the second row, such concepts and their cor-
responding neural representations are consistent across in-
dividuals from different translanguaging experiences when
 References
Alderson-Day, B., and Fernyhough, C. 2015. Inner speech:
development, cognitive functions, phenomenology, and neu-
robiology. Psychological Bulletin 141(5):931.
Boroditsky, L. 2001. Does language shape thought?: Man-
darin and english speakers’ conceptions of time. Cognitive
Psychology 43(1):1–22.
Chalmers, D. 1990. Why fodor and pylyshyn were wrong:
The simplest refutation. In Proceedings of the Twelfth An-
nual Conference of the Cognitive Science Society, Cam-
bridge, Mass, 340–347.
Chalmers, D. J. 1992. Syntactic transformations on dis-
tributed representations. In Connectionist Natural Language
Processing. Springer. 46–55.
Evans, N., and Levinson, S. C. 2009. The myth of language
universals: Language diversity and its importance for cogni-
tive science. Behavioral and Brain Sciences 32(5):429–448.
Fodor, J., and McLaughlin, B. P. 1990. Connectionism
and the problem of systematicity: Why smolensky’s solution
doesn’t work. Cognition 35(2):183–204.
Fodor, J. A.; Pylyshyn, Z. W.; et al. 1988. Connection-
ism and cognitive architecture: A critical analysis. Cognition
28(1-2):3–71.
Fodor, J. A. 1975. The language of thought, volume 5.
Harvard University Press.
Fodor, J. A. 2008. LOT 2: The language of thought revisited.
Oxford University Press on Demand.
Hussein, B. A.-S. 2012. The sapir-whorf hypothesis today.
Theory and Practice in Language Studies 2(3):642–646.
Kay, P., and Kempton, W. 1984. What is the sapir-whorf
hypothesis? American Anthropologist 86(1):65–79.
Markman, A. B., and Dietrich, E. 2000. In defense of rep-
resentation. Cognitive Psychology 40(2):138–171.
Niklasson, L. F., and Van Gelder, T. 1994. On being system-
atically connectionist. Mind and Language 9(3):288–302.
Rescorla, M. 2019. The language of thought hypoth-
esis. https://plato.stanford.edu/entries/language-thought/.
Accessed: 2021-01-12.
Wei, L. 2018. Translanguaging as a practical theory of lan-
guage. Applied Linguistics 39(1):9–30.