Overview
While rereading technical and popular publications on linguistics – this time closely – I realized that loose usage of terms language and speech, consciousness, thinking,
thought, semantics, meaning, etc. causes misunderstanding; writing a book titled What is language? might be a good idea. For now it is just a blog.
Language and speech
“To speak clearly and accurately we need a precise and well-defined language” [Aho1972::1]. Let us start with defining the terms language and speech.
The lack of clarity is apparent in distinguishing language and speech: speech act is typically separated from language, but speech as an object, a thing, a combination of signs: lexemes and phrases, a sentence – is not.
At best the two either used interchangeably or speech is considered a part of language [Cho1959].
These are the definitions of language and speech from the Random House Dictionary [Ran1996]:
language - 1) A body of words and the systems for their use common to people who are of the same community or nation, the same geographical area or the same cultural tradition. 2) communication by voice in the distinctively human manner, using arbitrary sounds in conversational ways with conventional meanings, speech. 3) the system of linguistic signs or symbols considered in the abstract (opposed to speech). 4) any set or system of such symbols as used in a more or less uniform fashion by number of people, who are thus enabled to communicate intelligibly with one another. 5) any system of formalized symbols, signs, sounds, gestures, or the like used or conceived as means of communicating thoughts, emotions, etc. 6) means of communication used by animals. …
speech - 1) the faculty or power of speaking; oral communication; ability to express one’s thoughts and emotions by speech sounds and gesture. 2) the act of speaking. 3) something that is spoken; an utterance, remark, or declaration. 4) a form of communication in spoken language, made by a speaker before an audience. 5) any single utterance of an actor in the course of a play, motion picture, etc. 6) the form of utterance characteristic of a particular people or region; a language or dialect.
I assume that the lexicographers at Random House did their job professionally and documented most frequently assigned meanings to the words in minds of English speakers: the job of lexicographers is to observe and document. Spreading clear and accurate definitions or meanings is the job of scientists, scholars, and science writers.
It is remarkable that in case of speech the definitions are not that far from the current view of majority of linguists (I hope), even though they are predominantly descriptions of speech act. I think that replacing “ability to express” by “expression of” in the third of the first three meanings will make the definition pretty accurate.
For language the situation is desperate. Skipping the detailed analysis of the consistency, accuracy, and precision of definitions, let us just note that they are mostly about “communication”; some of them about the “thing”, the others – “process”, and neither of them mentions anything close to grammatical rules. However, the first meaning can be made less vague by adding “the systems of rules for their use”, period.
From my far from comprehensive review of technical literature I found only one close to reality definition of language:
Socially shared code or conventional system for representing concepts through the use of arbitrary symbols and rule-governed combinations of those symbols. [Owe1988::453]
This is a very accurate definition of language: after removing the “Socially shared code or” from the beginning and adding “as a shared code for meaning” to the end it can be considered perfect. The speech definition, on the other hand, is not that consistent or accurate there:
speech - Dynamic neuromuscular process of producing speech sounds for communication; a verbal means of transmission. [Owe1988::456]
speech act - Basic unit of communication an intentional, verbally encoded message that includes the speaker’s intentions, the speaker’s meaning, the message’s meaning and the listener’s interpretation. Concept specified by Searle. [Owe1988::456]
Clear definition and demarcation of complex phenomena such as consciousness, thought, language, and speech is very important not only for speaking clearly: for the majority of us speaking is not the end. It is a means for expressing our perception of the universe, the reality and communicating it to others to understand.
Before getting into the definitions of language and speech it is important to mention that the term written speech in English technical literature is not quite common. However, the phrases oral speech, written speech, or gestural speech sound natural to Armenian or Russian linguists (sometimes text used for both - see, for example, #8 in Լեզվաբանություն եւ ՀԹՏ section of the Լեզվաբանություն, ԱԲ, եւ ՀԹՏ): see a definition of language at 4:40 and text - 6:25)
Has language anything to do with communication?
I would say No, but it depends on what one means by communication. See more details in the Is language a communication system? section below.
You can consider two humans exchanging verbal, written, or gestural messages encoded in particular language as a basic communication system involving speech. Their two minds are users or operators of the system: one (speaker’s mind) 1) generates thoughts, 2) using “language organ” encodes the thought into [written, vocal, or gestural] speech, while the other (listener’s mind) 3) parses (decodes) the speech (messages) using the same “language organ”, and 4) transforms it into a thought - understands. Here "same” means “same type” and similarly configured, because either brain contains distinct “language organ”.
In this scheme the phonemes, tonemes, graphemes, or gestemes are just physical representations of signs (symbols). The set of signs (symbols) and the rules for combining them into more complex signs that are known to both: the Speaker and the Listener. Knowing of sign includes knowing of respective signifiers – the meaning.
People with no impairment can use all types of them, but phonemes (tonemes) are basic. However, what is important is the sign, the symbol, rather than phoneme: we speak using symbols. It is interesting that writers do not pay much attention to this. They write sounds rather than symbols or signs – see, for example, [Cho1967::397], [Cha2018, in the subtitle and elsewhere in the text].
The difference between language and speech was first (to the best of my knowledge) pointed out by F. de Saussure. He underlined the diachronic nature of language and the synchronic – of speech [Cry2010::431, Ջահ1974::21], but by speech he meant speech act.
Diachrony is absolutive by nature, while synchrony – is relative.
"During application of language, i.e. during speech, synchronous units that are drawn into the process devoid time dimension. speech as a process expands over a duration of time. However, such dynamics should not be identified as history. From perspective of dynamic phenomenon speech has no history - it is a transient phenomenon in the true sense of the word, that is recordable and comparable to other instances of speech to the extent of getting recorded as language. The true bearer of history is language, the individual manifestations of which can be sequentially compared. At the same time, speech has a continuous character (pauses are also carriers of meaning and they enter speech as a necessary element), language is intermittent (discrete)" [Ջահ1974::21].
The units (signs, words) of language relate paradigmatically (associatively), while the relations of units (signs, words) of speech is syntagmatic [Cry2010::431] – linearized hierarchical, tree-like structure.
Chomsky who does not explicitly distinguish language and speech [Cho1959] “drew a fundamental distinction (similar to Saussure’s langue and parole) between a person’s knowledge of the rules of language and the actual use of that language in real situation. The first he referred to as competence; the second performance. Linguistics should be concerned with the study of competence, and not restrict itself to performance – something that was characteristic to previous linguistic studies in their reliance on samples (or ‘corpora’) of speech (e.g. in the form of a collection of tape recordings)” [Cry2010::433].
In this write up we will use these definitions of language and speech:
language – a set of [morphological and syntactic] rules that produce new signs or sequences of signs by combining them from previously available signs. Language is a speech producing system. [Հայ2022::282]
speech - combinations of signs of a certain language, produced according to grammatical rules, to convey information (content) from the speaker to the listener. Speech is the product of language. [Հայ2022::283]
"The distinction between language and speech has its parallel in the distinction between consciousness and thinking. Thus, we should talk not about the connection between language and thinking, but rather make parallels between language and consciousness and between speech and thinking respectively. Indeed, thinking, like speech, has a direct and processual character, while consciousness does not. It is nothing more then some existing state. Every act of thinking relies on consciousness and turns into a fact of consciousness. However, we should keep in mind that non-conscious phenomena (cf. interjections) also play an important role in language" [Ջահ1974::21-22].
Summary: language is a set of meaningful signs (terms) and a set of rules to combine them into new meaningful signs – the elements of speech (I cannot use the term Parts of Speech, because it is [not very accurately] assigned to represent linguistic, pertinent to language categories of signs like noun, adjective, etc.). Speech is a tree-like combination of signs that often times refer to some perceptions of reality. The relation between language and speech is a producer to product relation. Similar to: Ford Motors factory to Ford car; cookie cutter to cookie; printer to book, tree to fruit, C++ programming language to a program written in C++, etc.
Sense and Reference
F. de Saussure addressed another pair of linguistic notions: sign and its meaning. He “recognized two sides to the study of meaning, but emphasized that the relationship between the two is arbitrary (p.428). His labels for the two sides are signifiant (‘the thing that signifies’, or ‘sound image’) and signife (‘the thing signified, or ‘concept’). This relationship of signified to signifier Saussure calls a linguistic sign. The sign is a basic unit of communication within the community: langue is seen as a ‘system of signs’ ” [Cry2010::431].
At about the same time, a bit earlier [Fre1892], it turned out that the simple pairing is not quite accurate. It is unfortunate that linguistic community did not appreciate the work of Frege until now (at least in [Cry2010] there is no mention of Frege, while less accurate approach of Saussure’s is described).
"The late nineteenth-century mathematician-philosopher Gottlob Frege provided a concise distinction between these two often-confusing aspects of the word meaning. He distinguished between the sense of the term and its reference. Its sense is the idea that one has in mind that corresponds with considering a particular word or phrase. This is distinguished from the reference of the same word or phrase, which is something in the world which corresponds with this term and its sense" [Dea1998::61].
These developments clarified the process for communicating ideas from one mind into another. The specific language is a registry of categorized signs – morphemes - and their possible sense (meaning) and/or grammatical functions – a Dictionary. It is also a registry of combinatorial Rules – syntax - for encoding more complex notions, categories, [proper] names, phenomena that are missing in the Dictionary. The codes in the Dictionary and new codes make sense for a community of individuals that agreed 1) on the Dictionary codes and their mapping into things or concepts, as well as 2) on the Rules for encoding new things and concepts. The rules constrain the combination of categorized signs (parts of speech) and their derivatives (paradigmatic forms) into codes that make sense.
Morphological rules can be considered a part of syntax, because they also create new codes – lexemes, that typically called words - a very vague concept.
The codes that make sense we call grammatical or grammatically correct.
On the speaker’s side the function of language is completed at the point of creation of the code. For the listener the function of language starts at the moment of perceiving the speech and ends after parsing it into syntactic structure. The reference is assigned dynamically to syntactic structure after parsing. For listener it is not necessarily the same reference as for speaker. This is what causes misunderstanding.
The term dynamical means that the reference is not attached or linked to the code (the speech instance, the text) statically in the Dictionary or at the point of producing a new grammatical code. The meaning of morpheme or combination of morphemes depends also on the context they are in [Հայ2022::160-163]. You cannot build a grammatical sentence just using syntax. Typically it is a recursive process of 1) building a grammatical code, 2) calculating the reference, and then repeating 1) and 2) on the next level.
Language – more specifically the set of syntactic rules - is a tool for producing a speech, which makes sense. This is an independent value of syntax, the basis of its autonomy [Cho1975::138].
It is important to make clear the meaning of the term sense in the above context. This is Frege’s sense [Fre1892], rather then the everyday meaning: like in makes [no] sense. Frege’s makes sense is equivalent to understandable, to grammatical phrase. From that perspective the sentence "Colorless green ideas sleep furiously" [Cho1975::138] makes sense, but it does not have a reference - it is meaningless.
This linking is very much similar to the last link-edit step [Aho1972::721] of programming language compiler, when the parsed code, which was originaly written in high-level programming language, is edited to make it executable. We could call this Fregean binding.
Summary: the meaning has 2 aspects - sense and reference (see also Իմաստային պարադոքսներ). Sense is the content of a phrase, while reference is what does the content mean.
Words for concepts: encoding and decoding images
In everyday life we map words to things or to concepts loosely. In everyday life it turns out fine even in extremely ridiculous cases. For example, people perfectly understood that by social distance the health care authorities actually meant physical distance. Due to ‘smartness’ and flexibility of our thinking – but more precisely due to the late binding (Fregean binding) of speech - the majority of us kept physical distance (6 feet) and, in some degree, helped with getting recent pandemic under control.
In everyday life we get away just fine with using the word language for speech and vice versa. This ambiguity does not much affect our understanding of the world and each other. The context allows us to link it correctly on the fly to the right concept. When context is not enough the interlocutors can ask questions and enrich it.
Due to Fregean binding “inaccurate” phrases like part of speech, universal grammar, social distance, large language model, etc. acquire single clear meaning in particular context, at comprehension time. In these phrases some words (lexemes) have the expected binding – part, grammar, distance, large model. When less or no lexemes have expected binding and the unexpected, new binding gets stronger and stronger over time, we get an idiom.
Same can happen and happens to scientific terminology: when there is no consensus the binding process does not converge to single meaning. This is especially true for a complex, hard to define terms like consciousness, thought, thinking, language, speech, meaning, etc. It is very important to define such terms explicitly. A clear definition is important, because even if no one agrees with it, it will make discussion and argumentation meaningful.
Scientist use clear definitions of terms to set up a context that simplifies the binding of their speech to the description of reality, of the universe.
Politicians, on the other hand, consider multiple contexts: one that ensures a dynamic binding for supporters, another – for opponents, yet another - for ordinary people, who do not want to hear from them anymore. The best, the most skillful politicians produce such speeches that each audience segment unambiguously binds meaning according to the politician’s intend. The supporters are supposed to link such a meaning that they would not run away from politician’s party, the opponents – would have hard time finding objections and refutations, and the ordinary people – to think that the speaker’s goal is not for gaining power, but rather – finally - doing some good for them. This is a scientific explanation of doublespeak (see G. Orwell. Nineteen Eighty-Four).
Controversy in science means that the philosophers are in the process of defining/clarifying the terms, of binding sense to reference. I did not say that – Wittgenstein did (see, for example, ## 36, 43, 90, 118). He did it philologically, while I am trying to convey the same logically, technically by describing the exact place of misunderstanding using Frege’s theory of meaning.
Summary: In everyday speech we assign meaning to words loosely, which causes misunderstanding and requires additional interrogation. In scientific, technical literature terms must have clear, unambiguous meaning.
Linguistic determinism and relativity
Origins
The Linguistic determinism and relativity theory claims that 1) the human thinking, worldview is shaped by language (determinism) and 2) there are untranslatable statements (relativity).
This quote is considered as the definition of the basics of theory:
“We dissect nature along lines laid down by our native language. The categories and types that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, the world is presented in a kaleidoscope flux of impressions which has to be organized by our minds - and this means largely by the linguistic systems of our minds. We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way - agreement that holds throughout our speech community and is codified in the patterns of our language [...] all observers are not led by the same physical evidence to the same picture of the universe, unless their linguistic backgrounds are similar, or can in some way be calibrated” [Who1956::212-214].
This passage is quoted in [Bea1959::587], [Pin1994::49], [Cry2010::15]; the version above is from Wikipedia.
Analysis
I would argue – as many already did - that the opposite is true: language has nothing to do with the picture of universe. By just theorizing about cognitive pathways, using just clearly defined terms one can come to a very natural, even a trivial conclusion that thoughts shape the language.
“Humans parse images that come to the brain from senses into ontological, relational and logical concepts. The environment (in broader sense) is presented as relations of invented concepts, their breakdown and chunking. These make up the humans perception of the universe and of what they identify in their brain with reality. This is the picture of reality.
Modern science still does not know how this picture is formed and stored in the brain. People judge about that based on presenting and communicating the image of reality to each other. Communication is done by speech, be it natural (including mathematical and logical) or pictorial (drawing, picture) or another type.
Natural speech, together with the senses, conveys information about the environment. With this, it participates in the construction of the Semantic network (SN) [Hof1979::370-372].
Thus, the construction of SN is done by 1) cognitive and 2) evidential processes.
Senses ⇒ Image ⇒ ((Consciousness)) ⇒ [[Thinking]] ⇒ Concept ⇒ SN:
SNsp.Concept ⇒ ((language)) ⇒ [[Speaking]] ⇒ ((language)) ⇒ Concept ⇒ SN.
In the above bullet-items the actions are put into double square brackets, and the sets of laws or structures (as opposed to things) in double brackets. The speaker's SN is marked sp to distinguish the speaker’s SN from the listener’s. The first is a schematic description of thinking - the process of knowing, and the second - the communication process. With the latter, people report to each other what they saw, heard, thought about reality and events (see below for more details)” [Հայ2022::179 - sorry for self-quoting].
The intent of this quote is to show that Concepts and their relationships – the content of SN, which represents the reality in our brains - are formed upon data or information input from 5 senses and speech. You can consider the “language organ” 6-th sense. If eye conveys visual images to brain for converting them into Concepts, then “language organ” conveys speech to the brain. However, there is a big difference: the 5 sense are primary, immediate sources of images, while speech is secondary, mediated source – someone has already saw, heard, smelled, touched, or tasted something or imagined relevant image before converting it into Concept and then generating speech by “language organ”.
If we accept the above depiction of observing the world and talking about it, then we may notice that:
1. The set of concepts in individual mind can be created/updated by
a. observing (more generally – sensing and thinking) the world
b. listening to the description of concepts that are in the minds of other humans
2. Language cannot affect thinking because we do not think in natural, native language (see Do humans think in language? section below)
3. Any concept has to be formed first by sensing/thinking (in the ancestral mind).
4. In any monolingual community the thinking shapes language via:
a. Dictionary of morphemes denoting concepts,
b. Syntax by a process of grammaticalization [Cor2011::29-31] – roughly: a process of losing ontological or metaphysical sense and acquiring grammatical: genesis of function words, suffixes, adpositions, etc.
5. In monolingual society the only way that language can affect the content of the SN or affect cognition or worldview is inability to code a concept, to utter it.
Evidence (lack of)
The supporters of Linguistic relativity in 20s to 50s and recently were trying to find evidence for the bullet-item #5 above.
The two claims, which considered as evidence for Linguistic relativity: 1) the Eskimo language has about 100 words (started with 4 and then multiplied by journalists) for snow and 2) the Hopi language does not have tense, are not factual. See the details, for example, in [Pin1994::50-54], [Dev1997::108-112], [Prn2011::181-183].
“The empirical research discussed in previous sections (partly stimulated by Whorf’s own imaginative ideas) indicate that the cognitive processes studied so far are largely independent from peculiarities of any natural language and, in fact, that cognition can develop to a certain extend in the absence of knowledge of any language. The reverse does not hold true; the growth and development of language does appear to require a certain minimum state of maturity and specificity of cognition” [Len1967::364].
You need not to go into exotic languages to find out that some notions have no predefined sign (word) in a language. The whole purpose of the language is to provide tools for creating combinatorial signs for those. In the Indo-European family you can find languages that do not have words for maternal uncle, older brother, thirst, etc. But this does not mean that speakers of these languages cannot imagine – have concept of - maternal uncle, older brother, or never feel or think that they are thirsty.
Also it should not be surprising that Eskimo people who spend most of the time surrounded by snow, which gets different colors, textures, forms during year have many words for snow, while Maasai might have none.
Same is true for languages that do not have tenses or words for numbers. In such languages the grammaticalization of time related notions has not yet happened and might never happen; the numbers and counting are not necessary for happy life.
These cases can be explained by the environmental pressure (actually, the lack of): the ubiquitous principle of least effort.
Logic (lack of)
Suppose there is a botanist who looks at grown tree with oranges and asks: “Why everybody says that oranges grow on trees? Let us revolutionize botany. Let us imagine that oranges produce the tree. First, imagine that there are oranges in same spots, as they are now, with no tree. The orange spheres will look beautiful on the background of this deep blue sky with small patches of bright white clouds. What if all of a sudden they start growing branches with dark green leaves that start merging and heading towards ground? And finally, these branches combine into a healthy, heavy trunk; the gravitational force then hammers it deep into the soil. Isn’t this beautiful? The background, the colors, the shapes. … I need to put this down: ‘We dissect nature along lines …’.”
Then, a new generation of botanists come into scene and point out that the revolution got out hands and it is not quite right: where from the oranges appear in the sky in the first place? “But look” they say, “the oranges do shape the tree. They bent the branches beautifully into perfect arches… and the heavy ones even fractured them, … amazing”
Do the Whorfians mean this kind of bending when they say that language affects thinking? We do not know, because no one explains what they mean by language, thinking, thought (whose: the speaker’s or the listener’s?), shaping, etc.
Two aspects of shaping the tree by oranges are important: 1) the oranges change the appearance of the tree, rather than the essence, 2) the oranges shape the tree not by virtue of their essence, not because of they are oranges, but rather because they are heavy: bricks hanging on a tree might have the same or better effect.
“To find evidence for the Sapir-Whorf hypothesis, we need three things: a well-established linguistic difference between two languages, a correlation between that difference and psychological difference between speakers of those languages, and a reason for thinking the latter caused by the former” [Pri2012].
I do not want to nitpick on these 3 things - they are absolutely necessary. But are they sufficient? I think that it is also absolutely necessary the fourth thing: a reason for thinking that there are no other differences that “cause the former”. For the orange tree metaphor you need to establish that only oranges are bending the branches: there are no bricks, no apples, no car parts, no pieces of clothing, etc. that can do the same. In this case you can be sure that the oranges, rather than weight, do the bending.
The logic of Whorf’s theory goes like this:
1. God put languages into the humans brains:
a. “Now the whole world had one language and a common speech.”
b. “If as one people speaking the same language they have begun to do this, then nothing they plan to do will be impossible for them. Come, let us go down and confuse their language so they will not understand each other.”
2. Kind, wise, and almighty God created words like baby, daddy, granny, kitty, mommy, puppy for people to adore and love babies, daddies, grannies, kitties, mommies, puppies
3. If God had not been kind and wise, she/he would have named them biden, erdogan, osama, prigozhin, putin, xi and we would hate them.
Summary: The facts and experiments so far point towards the conclusion that thinking shapes language (rather environment shapes the depth and width of grammaticalization).
What is language not
Is language a communication system?
If we accept the definition of the language given in this write up then it is not. To give a more detailed and accurate answer we need first to define communications system. You can name a system for conversion thoughts into speech a communication system: language in this case communicates (converts) thoughts to speech and speech to thoughts.
The goal of communication system is to transfer information (news) via messages from a source to target, from an emitter to the receiver of information. It consists of a 1) source, 2) transmitter, 3) channel, 4) receiver, and 5) destination [Sha1949].
Language is a tool for producing (encoding) and consuming (decoding) messages that contain human thoughts.
Different communication systems: telephone, radio, tête-à-tête [conversation], writer - hand and stylus - tablet - eyes - reader, speaker-mouth (and some surrounding organs) - air - ear - listener.
Do animals have language?
If we accept the definition of the language in this write up then they do not. Language is a system of signs and rules for dynamically encoding/decoding unbound number of thoughts into new signs. Animal communication or, more accurately, signaling system is based on hardcoded, static [Cry2010::421] mapping of signals to actions – a lookup table. They cannot combine new signals to new actions – they cannot speak. Moreover, we cannot teach them to speak our language, because they lack biological tools, organs for that. This is not about organs to instantiate speech via gesture or sound, but about the “language organ” to produce and parse speech. They can perform speech, but they are not competent in generating (producing) or comprehending (parsing) it. Even the bee dance is genetically hardcoded [Cry2010::421] parameterized signaling.
For the same reason of not having a suitable organ – the human shoulder – other animals cannot be taught to throw things (stones) by limbs.
Do humans think in [natural] language?
No. See [Len1967::364] quote in the Linguistic determinism and relativity section above and the Mentalese section [Pin1994] for more detailed analysis. The conclusions in the latter are: 1) thoughts are sharp [AH: they have single peak, bell-shape distributions], words are fuzzy [AH: they have multi-peak (multiple meanings) bell-shape, chains of peaks distributions], ambiguous; 2) the logic of thoughts is sharp, [English] language lacks logical explicitness [AH: I doubt it; the examples [Pin1994::70] are very concerning, very strange]; 3) co-reference (e.g. John, the man, him can refer to the same person); 4) contextual interpretation of deixis [AH: Why not “every linguistic unit”?]; 5) synonymy: single event can be described by multiple sentences (e.g. actively and passively).
Why GPT is not a language model?
First we need to agree on what is a model. Not formally, simply put: it is a simplified copy of things, objects, or phenomena.
GPT is a neural network trained to continue a string of characters according to statistics gathered on large amount of texts from the Internet. It is an “autocomplete on steroids” [Debunking the great AI lie | Noam Chomsky, Gary Marcus, Jeremy Kahn::15:59].
If you train the model by feeding strings of pixels associated with strings of characters, describing the picture, then it will start describing the scenes you submit. Or vice versa: if you feed it with the description of scenes associated with the pixel representation of the scene, it will start “painting” at your command.
Everyone will probably agree that neural network is a brain model. This is the first, basic layer of the model. On top of it the second layer is built. a predictive statistical model, which assumes that speech can be generated by calculating the next character (or word) based on some number of preceding characters (or words). It is important to note that this is probably not how humans generate speech. We assume that there is an innate “language organ” that builds, generates speech according to syntactic rules or constraints. Per currently accepted model humans build the utterances recursively as a tree-like structure and then linearize it. From that perspective GPT is a model of speech generator, rather than language, because it "does not know" grammar, it does not model competence - a technical term that “refers to the ability of the idealized speaker-listener to associate sounds and meanings strictly in accordance with the rules of his language” [Cho1967::398]. It models performance – the production and consumption of speech.
But here is an interesting question: is it possible that language model– the third layer - emerges as a result of statistical learning?
The language innateness hypothesis has a plethora of experimental support, but no hard evidence – no one identified the “language organ” in the brain. Maybe the “language organ” is a product of statistical learning by babies? Maybe baby-humans brain is designed to create the “language organ” on a very limited, poor input. The poverty of input or stimulus [Prn2011::144-157] is one of the major arguments for innateness.
Neural networks can become useful tools for investigation of human brain. How it learns and how it uses acquired “knowledge” [Հայ2022::179]?
There is new, emerging area of research: Mechanical interpretability [The Impact of chatGPT talks (2023) - Prof. Max Tegmark (MIT)::4:35]. GPT for these researches is what LHC (Large Hadron Collider) for physicists. Another interesting talk is the "Dr Stephen Wolfram says THIS about ChatGPT, Natural Language and Physics."
While conducting such a research, we need to keep in mind that we are dealing with a very specific learning: converting one string of numbers into another. If I take time and meticulously, pixel-by-pixel copy the Mona Lisa, can I claim that I am a painter or even more – Leonardo Da Vinci?
On the other hand the research in area of Mechanical interpretability can shed some light on language acquisition.
The possible experiment outline could look like this:
1. Identify phrases that people produce around an infant until a year of age.
2. Feed neural network – seq2seq (you can find the description in [Հայ2022::263-274]) or Transformer architecture - with those and then prompt the network and record results.
3. Add more phrases to the training sets, for example, adding the ones that baby can hear by 1.5 years of age.
4. Prompt the network and record results.
The last two steps can be continued with gradually adding phrases for 2, 3, etc. years old. Then by analyzing the prompt results after each training session, we can find out if the network starts “syllable babbling”, “gibberish babbling”, one-word utterances, two-word strings, and finally full sentences [Pin1994::273]. Using tools and methods of Mechanical interpretability we can monitor if “interesting” clusters of neurons emerge during deep learning. Are there firing patterns during speech input and output identifiable with operations of parsing and generating speech. In other words is there evidence that the “language organ” or competence emerges in statistical learning.
It is safe to assume the poverty of Armenian input to GPT. We can have a glimpse of possible behavior for the suggested experiment by reviewing GPT answer to the last question on the Interviewing chatGPT: Armenian Corpora and morphology test page.
The Language evolution
[This section is inspired by the question to Dr. N.Chomsky in the #10.f video (00:14:38, Are there units of culture, like memes?) - see Լեզվաբանություն, ԱԲ, եւ ՀԹՏ).]
I am not going to talk about language evolution per se - I'll try to give an outline of possible approach. Chomsky's response just triggered an urge to explain what evolution and, in particular, "language evolution" is (as a part of cultural evolution).
There are a lot of misconceptions and misunderstandings surrounding evolution as a process. Despite clear definition and beautiful explanation by Darwin more than 165 years ago and neo-darwinists (e.g. R. Dawkins) currently, the biological evolution still causes confusion, misinterpretations, and rebuttal using the word of God, trickery, mystery, and magic.
Digression (sorry): Recently, scientific misinterpretations come mostly from the intelligent design community. It is very painful to view (search on YouTube for "intelligent design" or better yet "scientist proves that God exists") people with science degrees seriously debating complete nonsense. They like to prove mathematical impossibility of Darwinian evolution. Astonishingly, trained mathematicians exhibit elementary misunderstanding of the probability theory. [Digression in Digression; After writing "Astonishingly" I realised that it should not astonish anyone anymore - I highly recommend reading the Probability and Ambiguity section in [Gar2001] to understand why: “Charles Sanders Peirce once observed that in no other branch of mathematics is it so easy for experts to blunder as in probability theory” [Gar2001, p.273]. Honest, well trained mathematicians can honestly blunder. Here, in the upper Digression, I am not talking about them. End of DiD.] My "favorite" example of the intelligent design justification by improbability argument is citing Sir Fred Hoyle: "The chance that higher life forms might have emerged in this way is comparable to the chance that a tornado sweeping through a junkyard might assemble a Boeing 747 from the materials therein." This sentence may impress a laymen as a journalistic brilliancy during political debate, but it has nothing to do with science or probability theory application. There is a detailed analysis of the Junkyard tornado fallacy. I just want to reiterate the high-level problem in all known so far probabilistic refutations of Darwinian evolution (to be fair: Sir Fred Hoyle did not use the phrase for refuting evolution; he was talking about the origin of life on Earth). The general schema is like this: 1) a goal is assigned to a random process and then 2) it is shown that the goal could not be accomplished by a random process in a given timeframe - during the existence of Earth or the Universe. The goal cannot be correctly specified or identified in the first place (that is why no one so far did), and secondly, evolution has no goal. For clear explanation how randomness works in the evolutionary process see the Weasel program. End of Digression.
Let's first clarify some termes to not deepen the confusion.
It is very hard (to me) to define what is change, because it is one of the most mysterious and fundamental things that since the times of Parmenides we do not have deep understanding of. Here I'll stick with the common sense, with our everyday intuitive understanding of it.
There are 2 types of change: evolution - a gradual, "baby-step" change, and revolution - an abrupt and violent change, typically to the opposite (hence the re-).
There are 3 major realms of evolution, which is at times turns into a revolution: 1) the whole Universe and the Earth as product and a participant, 2) Earth's biosphere, and 3) Earth's, more precisely human's noosphere. Simply put the last 2 are biological and cultural evolutions respectively.
Writing a book "What's evolution?" is on my bucket list (-:), but not in the top 10.
With big brush strokes I would like to provide some perspective on understanding of the misunderstandings.
So, evolution is a gradual change. The speed of evolution of different realms is significantly different: for Earth - eons, for life - milliard years, for culture 10-th of thousands years. That why we should not talk about things like biological and cultural coevolution. For clarity we need to assume that bioevolution happens in the instance (blink of eye) of Universe evolution, and cultural in the instance of biological. From the cultural evolution timescale there is no evolution in life forms, from the biological evolution perspective - there are no changes in the Universe. When we consider bioevolution the Universe is static, when we consider cultural evolution - the life is static.
Chomsky shares this point of view when he says that language is not evolving. What he means by that is that the "language organ", the biological faculty that supports human linguistic ability (competence), the language [deep] structure does not change.
I hope that I also understand why he says that language changes rather than evolves. Biological evolution has a clear unit that replicates and evolves - the gene. Neither Universe, nor Culture does not have that. All the wonderful concepts like meme [Daw2005::189-201] or culturgen [Wil2006::238-274] do not go beyond being just a convenient term, a name. They do not reference any meaningful, self-sufficient essence. Unlike genes these concepts do not add to our understanding of cultural evolution; they do not provide explanations of things that we explain without them. Genes - long before discovery of DNA structure - explained, in particular, the distribution of inherent attributes.
Genes composition, structure, and replication mechanism is common for all life forms. What is common for memes in painting, growing and smoking tobacco, telling a joke, or hunting armadillos, making chairs, and believing in God? Design patterns popular in engineering and architecture [Ale1977] might be a good starting point for the quest.
["What is meme or culturgen?" is also on the list of books (-:), but this one is not even in top 20.]
Now back to evolution of language. Evolution definitely takes place and language has some kind of "genes" (or DNA). Imagine all lexemes of the language classified and represented by morphemes, all phonological and alteration rules, all combinatorial (grammatical) rules categorised and written in one string in BNF or any notation. This is the language DNA. Elements of speech are "proteins" out of which we build organisms - texts, stories, monologue, dialogs, broadcasts, etc.
There are also rules and processes that ensure 1) permanence - slow change of lexicon and grammar, 2) fecundity - copying of lang-genes is exponential (as number of human brains) : babies learn early and fast (adults are less successfully), 3) high-fidelity copying - when we learn a language we are forced to copy the lexicon and rules accurately, because no one will understand us. These attributes of a successful replicator [Daw2005::18, 193-195] along with lang-genes variability (mutation) are present in language. It is important to note the "high-fidelity copying" - if it were "absolute-fidelity copying" we will have no evolution. Evolution exists due to balancing permanence and change: with lower-fidelity there is no permanence, with higher-fidelity - no flexibility, no trying something new and potentially - better.
All of these needs a careful conceptualization and research to become a useful theory for investigating the "speech"-organisms. So far no concepts of language evolution have been suggested to increase explanatory power of linguistics.
A linguistics "genome" - langnome - can be build similar to suggested in [Հայ2022] for Eastern Armenian. It will contain language DNAs for all existing currently and before dialects.
We are building isolated and partial linguistic "proteomes" in corpora. But they do not isolate the linguistic "proteins" - Context Trees - as suggested in [Հայ2022].
Language studies and linguistics
Linguistics is a part of language sciences, which include:
1. Theoretical linguistics that is concerned with:
a. Common linguistics: typology, paradigmatic systems, grammar and grammar description metalanguage (inaccurately named Universal Grammar), universal deterministic language models, etc.
b. Grammar of particular language, dialect; orthography; deterministic special language models.
c. Comparative linguistics; historical linguistics; evolution and classification of languages, comparison of deterministic special language models.
d. Phonology
2. Computational linguistics
a. Corpus design; stemming, tagging, lemmatization
b. Methodology for statistical research
c. Statistical modeling of language
3. Experimental linguistics
a. Lexicography.
b. Etymology.
c. Corpora enrichment: sampling oral and written speech.
4. Philology
a. Storytelling forms; history and evolution of storytelling forms.
b. Corpora enrichment: sampling literature and folklore.
c. Pragmatics - this might be weel classified as a part of theoretical linguistics or, rather, a separate part on that level. Another possibility is - part of semantics (or synonym of semantics)
5. Psycholinguistics
a. Language acquisition; teaching
b. Language competence (speech generation and comprehension)
6. Neurolinguistics
a. Brain activity at speech generation and comprehension
b. Linguistic deficits
There are well established theories: theory of formal grammars [Гро1971], [Cho1959], [Cho1967] to mention very few, theory of compilation [Aho1972], Frege, Peirce, Russell, Tarski theory of sense, etc. What is missing in theoretical linguistics is a theory of reference, which will consider several levels of context [Հայ2022::161] for clarifying sense and calculating the reference.
Summary
“To speak clearly and accurately we need a precise and well-defined language” [Aho1972::1]. This will drastically improve mutual understanding and eliminate misunderstanding in scientific debates. The linguistics community might never accept single interpretation of consciousness and thought, language and speech, producer and product. If we acknowledge this situations and clarify our positions via definitions, at very least we will get a chance to be understood. Acceptance is a completely different and extremely complicated issue.
Sloppiness in definitions and thoughts expression in linguistic (actually any) contents look like the major cause of misunderstanding. The inability to express and unwillingness to understand the expression causes emotional explosions, which end with classifying certain theories as "conventional absurdity" [Pin1994::47] or breaking into argumentum ad hominem.
Mini review of [Cry2010]. It is strange that in [Cry2010] there is no mention that invalid evidence/facts were used to backup the theory of Linguistic relativity. [It is worth researching if Linguistic Relativity was taken seriously in continental Europe. It looks like that this kind of theories (another example is the theory of Intelligent Design) are pertinent to North America.] The major theoretical achievement of Panini, Stoics, and Port Royal are mentioned cursory, in passing.
There is no mention of Frege, who explained (or reminded) the two facets of meaning.
If we accept the tenets of this write-up then 1) thinking is not performed in any particular language, 2) language is a toolset for encoding thoughts into speech and decoding the speech into thoughts.
Theoretical linguistics is a hard science very much similar to physics. That is why linguistic theories are tentative, temporary.
References
[Aho1972] A.V. Aho, J.D. Ullman. The Theory of Parsing, Translation, and Compiling. V. I: Parsing. Prentice-Hall, Inc., Englewood Cliffs, N.J. 1972.
[Ale1977] C. Alexander, S. Ishikawa, M. Silverstein, M. Jacobson, I. Fiksdahl-King, S. Angel. A Pattern Language. Oxford University Press, 1977.
[Bea1959] R.L.Beals, H.Hoijer. Introduction to Anthropology. The MacMillan Company. NY. 1959.
[Bob2006] S. Bobzien. Ancient Logic. In Stanford Encyclopedia of Philosophy (online). 2006. [Cha2018] W. Chafe. Thought-Based Linguistics; How Languages Turn Thoughts into Sounds. Cambridge University Press, 2018
[Cho1959] N. Chomsky. On certain formal properties of grammars. Information and Control. 2 (2): 137–167, 1959
[Cho1967] N. Chomsky. The formal nature of language, Appendix A in [Len1967].
[Cry2010] D. Crystal. The Cambridge Encyclopedia of the English language, 3rd edition. Cambridge University Press, 2010.
[Cor2011] M.C. Corballis. The Recursive Mind. Princeton Univ. Press, Princeton and Oxford. 2011.
[Daw2005] R. Dawkins. The Selfish Gene (30th anniversary ed.). Oxford University Press. 2005.
[Dev1997] K.Devlin. Goodbye, Descartes. Wiley, John & Sons. NY, 1997.
[Gar2001] Gardner, M. The Colossal Book of Mathematics. W.W. Norton & Company, NY,. 2001.
[Jac1992] R.S.S. Jackendoff. Semantic Structures (Current Studies in Linguistics. MIT Press. Cambridge, 1992
[Fre1892] Über Sinn und Bedeutung (On Sense and Reference), Zeitschrift für Philosophie und philosophische Kritik, vol. 100, 1892
[Hof1979] D.R. Hofstadter. Gödel, Escher, Bach: An Eternal Golden Braid. (20th anniversary edition) Basic Books, 1999.
[Len1967] E. H. Lenneberg. Biological foundations of language. Wiley, NY, 1967.
[Lav2009] J. Laver. The Phonetic Description of Voice Quality (Cambridge Studies in Linguistics, Series Number 31), Digital printing, Cambridge University Press, L. 1980
[Owe1988] R.E. Owens, Jr. Language Development. An introduction. 2-nd ed. Merrill Publishing Company. Columbus. 1988
[Pin1994] S. Pinker. The Language Instinct. William Morrow & Co (HarperCollins Publishing) London, 1994.
[Pri2012] J.J. Printz. Beyond Human Nature. How Culture and Experience Shape the Human Mind. W. W. Norton & Company. NY. 2012
[Ran1996] Random House Compact Unabridged Dictionary. Special 2-nd ed. Random House. NY. 1996
[Rei1947] H. Reichenbach. Elements of symbolic logic. Macmillan. L. 1947.
[Sha1949] C. E.Shannon, W. Weaver. The Mathematical Theory of Communication. University of Illinois Press. 1964
[Who1956] B.L. Whorf. Language, Thought, and Reality: Selected Writings of Benjamin Lee Whorf (edited by J. B. Carroll). Cambridge, MIT Press 1956.
[Wil2006] E.O. Wilson. Nature Revealed. Selected Writings 1949-2006 (A summary of the Genes, Mind and Culture (by C.J.Lumsden and E.O. Wilson. Cambridge, Mass., Harvard University Press, 1981) with critic is in section Précis Genes, Mind and Culture). John Hopkins University Press, Baltimore. 2006.
[Гро1971] М. Гросс, А. Лантен. Теория формальных грамматик. Мир. М. 1971.
[Հայ2022] Ա. Հայրապետյան. Բնական խոսքի ընդհանրական ներկայացման մի տարբերակի մասին. Agoulis, Concord, 2022
[Ջահ1974] Գ. Ջահուկյան. Ժամանակակից հայերենի տեսության հիմունքները. ՀՍՍՀ գիտությունների ակադեմիայի հրատարակչություն, Ե., 1974:
コメント