ISSN 01476882, Scientific and Technical Information Processing, 2015, Vol. 42, No. 5, pp. 307–312. © Allerton Press, Inc., 2015. Original Russian Text © O.P. Kuznetsov, 2014, published in Iskusstvennyi Intellekt i Prinyatie Reshenii, 2014, No. 3, pp. 32–39.
Conceptual Semantics O. P. Kuznetsov Institute of Control Sciences, Russian Academy of Sciences, ul. Profsoyuznaya, 65 Moscow, 117997 Russia email:
[email protected] ...even tending toward abstraction, you can not come to terms with the loss of visibility, and the presentation that lacks the props of strong, stout sensuality, full of formulas, speaking about the stone more than the very stone being seen, experienced as taste and touch, such presentation is boring and burdensome for you, or at least, leaves a feeling of dissatisfaction, familiar even to outstanding theorists, your most fascinating abstractors. Stanislaw Lem, Golem XIV Abstract—The basic provisions of conceptual semantics that were proposed by the wellknown linguist S. Pinker are discussed in this paper. As well as other concepts of cognitive semantics, this concept empha sizes the fundamental role of human physical in the formation of cognitive processes. One important thesis of S. Pinker is the assertion of the existence of a language of thought, which serves to represent word meanings in the mind and is primary in nature relationship to verbal language. Keywords: conceptual semantics, language of thought, cognitive models, discourse DOI: 10.3103/S0147688215050044
INTRODUCTION One of the fundamental concepts of modern com puter science is that of formalization, which appeared long before computer science and even computers. It came from research on the foundations of mathemat ics, which began in the second half of the 19th century and was aimed at an increase in the severity of mathe matical proofs. The solution of a formalized problem should reduced to purely formal operations over its be description. The idea of a formalized approach was to exclude from the process of solution (or theorem proofs) such subjective features of the solver as knowl edge, experience, and skill that differ from person to person, and thereby to unify the decision process as much as possible. In the 1930s this approach led to the idea that the decision process should be a mechanical procedure, which can be performed by a machine. Thus, by the time of the first computers the conceptual base for its creation and use was fully prepared. The formal approach does not mean the lack of consideration of the content of a problem and its meaning. However, it clearly divides syntax (i.e., rules for the construction of language expressions in the problem descriptions and operations over it) and semantics, i.e., the meaning of language expressions. Moreover, it claims that the solution can be carried out by purely syntactic methods; semantics occurs only at
the beginning (in the problem statement) and at the end, with understanding of the formal solution. In other words, syntax exists independently of semantics in the form of a formal system or (equivalently) formal language. The first successful realization of this approach was carried out in logical calculi (formal theories). In A. Tarski’s works [1] the relationships between syntax and semantics in formal theories were clearly formu lated with the precise concept of interpretation as an unambiguous mapping of the elements in formal lan guage onto the elements of the subject area, corre sponding to the contents, i.e., theory semantics. This was a success due to the fact that the usual mathemat ical theories with wellstructured structures were sub ject areas for which there were formal theories. The formalization of these theories consisted in their axi omatization, i.e., in their transformation into logical calculi. A classic example of such an axiomatized the ory is formal arithmetic. The second successful realization of the formal approach, which played a predominant role in the computerization of various human and social fields, consisted of programming automation facilities based on universal programming languages that appeared in the 1950s. The interpretation procedures in the sense of A. Tarski are translation procedures and the seman tics are programs in machine codes.
307
308
KUZNETSOV
At the same time, a theory of formal languages and grammars was developed by the linguist N. Chomsky, which was a theoretical generalization of the descrip tion of programming languages (Backus–Naur form). This theory, having played a significant role in the cre ation of the mathematical foundations of structural (and subsequently computational) linguistics, pro vided great hope for the possibility of constructing a formal model of natural languages, which, in particu lar, could solve the problem of machine translation by purely syntactic means. The reality was, however, much more difficult. The inadequacy of the formal approach was soon revealed, as described in the review [2] (which quoted [3]): “in the course of the construction of a formal algorithmic language model, the algorithm and rules become com plicated at each stage. Despite this fact, each time it turns out that some fragments of natural language are not generated and, conversely, the things that are gen erated do not exist in this language; thus, new improvements are needed. Therefore, the framework of the narrow algorithmic model allows us to ask a question that in the human manner of thinking and speaking is not algorithmic.” This conclusion means the following. The proce dures of the formal (algorithmic) approach are mech anistic and do not depend on who executes them, a human or a computer. If people think in a manner that is not algorithmic, the properties of natural language must be associated with the peculiarities of cognitive processes. Thus, cognitive linguistics came into exist ence, from which cognitive semantics occurred and began to play its own role. The Cognitive Semantics of J. Lakoff Syntax does not play a predominant role in the human cognitive system. Compared to the computer and formal systems, people do not use insignificant symbols. “When a person proves a statement, he relies on its meaning, without building a formal proof. A reasonable person who did not study logic is capable of being understood exactly and identifying contradic tions quickly, but cannot often construct a formal proof” [4]. The most detailed concept of cognitive semantics was suggested by J. Lakoff [5]. Its basic principles are described in [3]. Here, we will formulate them as fol lows. 1. Values arise earlier than conceptual structures: they arise from our preconceptual bodily experience. Our experience is preconceptually structured at a basic level. The preconceptual structures are Gestalts and imageschematic structures generated by our experi ence, such as container, up–down, part–whole, and center–periphery. 2. The basic level of concepts (the prototype level in E. Rosch’s concept [6]) is in the middle of the hierar
chy from the abstract to the concrete and is character ized by the presence of a holistic and visual mental image, viz., Gestalt. 3. The imageschematic structures are the basis of human reasoning, which largely replace the rules of logical inference. In particular, the “container” schema is an image analog of Modus Ponens: If con tainer A is in container B and X is in A, then X is in B. 4. The understanding problem is formulated in terms of gestalt and imageschematic structures; it is extremely essential when describing cognitive pro cesses, but essentially inexpressible in the formal approach. It can be said that formal reasoning is “rea soning without understanding.” It is obvious that the problems that are raised by cognitive semantics go beyond actual linguistics and are of great interest for artificial intelligence. Here, we will consider an additional concept of cognitive semantics that was presented in a book [7] by Stephen Pinker. The Conceptual Semantics of S. Pinker is the Language of Thought that Precedes Verbal Language Pinker clearly formulated the difference of cogni tive semantics from formal semantics in the sense of A. Tarski at the very beginning of [7]. If formal seman tics is the relationship of formal language to a subject area, then cognitive semantics is a set of relationships. “Semantics is about the relationship of words to thoughts, but it also about the relationship of words to other human concerns. Semantics is about the rela tionship of words to reality: the way that speakers com mit themselves to a shared understanding of the truth, and the way their thoughts are anchored to things and situations in the world. It is about the relationship of words to a community, how a new word, which arises in an act of creation by a single speaker, comes to evoke the same idea in the rest of a population, so peo ple can understand one another when they use it. It is about the relationship of words to emotions: the way in which words don’t just point to things but are saturated with feelings, which can endow the words with a sense of magic, taboo, and sin. As well, it is about words and social relationships: how people use language not just to transfer ideas from head to head but to negotiate the kind of relationship they wish to have with their con versational partner” [7]. One of the main theses of Pinker’s concept, which he calls conceptual semantics, is that “word meanings are represented in the mind as assemblies of basic con cepts in the language of thought.” In other words, the language of thought is the language of semantics that precedes verbal language. The concepts are stored in a form that is far more abstract than concrete sentences. It is confirmed, in particular, by the fact that people remember concrete sentences much worse than the meaning that they have learned from them (and which
SCIENTIFIC AND TECHNICAL INFORMATION PROCESSING
Vol. 42
No. 5
2015
CONCEPTUAL SEMANTICS
can be different from the meaning that the speaker intended). It is necessary to note that this thesis directly con tradicts the wellknown Sapir–Whorf hypothesis, which states that the language structure determines thinking and the manner of understanding reality. The basic components of the language of thought of Pinker are a set of abstract conceptual frameworks that organize our experience, viz., space, time, sub stance, causation, force, and number. More specif ically, the composition of these components is as follows: • A system of basic concepts: event, state, thing, path, place, and property; • A system of relationships that connect these con cepts: action, motion, existence, and possession; • A system of spatial concepts to define a place and path that is expressed in prepositions as “on,” “at,” “in,” “to,” “over,” and “under”; • A time line that orders events and that differenti ates instantaneous events, bounded intervals, and has indefinite regions; • A set of causal relationships: causing, letting, enabling, preventing, impeding, and encouraging; • The concept of a goal, and the distinction between the means and the goal. It should be noted that the existence of innate con cepts and relationships are recognized by many cogni tive researchers, although the specific lists of different authors are different: “Basic innate concepts–primitives are reduced, as far as is now known, to a list of about 30 units that are associated with space and movement in it: the begin ning of a “path” and the end of a “path”; inside a “container” and from a “container”; on the surface and from the surface; up and down; connection; con tact; rhythmic/intermittent motion and similar motion; live objects begin to move without external influences (connections and contacts) and rhythmi cally; inanimate objects whose movements require external influences, etc.” [8]. However, in terms of conceptual semantics the issue is not just the innateness of matter, i.e., in the sequence of the appearance of the various elements of linguistic structures in the ontogeny of thought and speech. The fact is that the primacy of the “language of thought” is preserved through the entire human life and, therefore, determines both the representations of information in the mind and the mechanisms of its verbalization. Two statements Einstein made are a good illustra tion of this thesis. One of them was expressed in a con versation with Wertheimer: “I very rarely think in words at all. A thought comes, and I may try to express it in words afterwards” [9]. Another is contained in Hadamard’s remarkable book [10]: “Words, as they are written or spoken, do
309
not seem to play any role in my mechanism of thought. The psychical entities that seem to serve as elements in thought are certain signs and more or less clear images which can be “voluntarily” reproduced and com bined. … The abovementioned elements are, in my case, of a visual and some of a muscular type. Conven tional words or other signs have to be sought for labo riously only in a secondary stage, when the mentioned associative play is sufficiently established and can be reproduced at will.” Thought and Language Structure the World The second important thesis of Pinker’s concept is as follows: thought and language impose the structure of the world on us. “Our cognitive models represent the data that are read from the main aspects of human nature. Each of these models of understanding corre sponds to clearly human purposes and they allow us to distinguish substance, space, time, and causality, as it is primarily important for the realization of human 1
goals in nature and society.” This thesis has nothing to do with the Sapir–Whorf hypothesis, on which, as already mentioned, Pinker has a negative opinion. Strictly speaking, the world structures thought and language provides the means for structuring. Pinker’s book [7] contains very exten sive linguistic material that shows the variety of these means. Human goals at different times are different and the methods of the “division of reality” have to be dif ferent. This corresponds to the ability of thought “to structure even the most plodding everyday events in more than one way,” which occurs by various linguistic means. For example, we say that hay is loaded into a wagon (an impact on hay), but we can also say that the wagon is loaded with hay (an impact on the wagon). Pinker calls this restructuring a gestalt shift, which is similar to “the classical case of the “face–vase” visual illusion in which the figure and ground switch places.” Thus, it does not depend on the actual physical prop erties of an object or process and is generally deter mined by the goals of the speaker. Types of restructuring vary. Both a change of the affected entity (hay–wagon), and the “holistic effect” occur: the association of an object that is set in one object (pebbles–gravel or people–crowd). Often the restructuring can be called “framing”: the inclusion of the object/process in some frame, whose choice can strongly change either the pattern or its evaluation. “Many disagreements in human affairs turn not on differences in data or logic but because of how a prob lem is framed (invading Iraq or liberating Iraq, etc.).” The cognitive models of a person are his common sense and everyday thought. Our mind creates our 1 In
the future, all quoted texts are quotations from the Russian translation of [7], unless otherwise stated.
SCIENTIFIC AND TECHNICAL INFORMATION PROCESSING
Vol. 42
No. 5
2015
310
KUZNETSOV
worldview and this picture (intuitive geometry and intuitive physics) can be very different from what the world actually is, i.e., how it is presented by scientific knowledge. The simplest examples are wellstudied visual illu sions. The adult brain learns to deal with them on the basis of experience. The description of space in our language is very different from the representations of geometry. According to geometry a threedimensional body is limited by a twodimensional surface, a sur face, by a onedimensional line, and a line, by a zero dimensional point. However, the word “end” can mean the short line that limits a tape and the two dimensional surface that limits a whetstone. Restruc turing occurs here again. At the moment when we speak about the end of a tape, a whetstone, or a road, our thought structures them as onedimensional objects, ignoring their other dimensions, although in other situations their width is also important. “Few people think of a wire as a very, very skinny cylinder and of a CD as a very short one, though tech nically that is what they are. We conceive of them as having only one or two primary dimensions, respec tively.” “Why do we say that something can be underwater or underground even though it is surrounded by, and not beneath, the water or the ground, but is not under them? It is because water and ground are conceived as twodimensional surfaces, not threedimensional vol umes, geologically improbable though that is.” The “end” concept can belong not only to objects that are located in space but also to processes that are located in time: we speak about the end of the working day, house cleaning, and a football match. This is because time is represented by spatial metaphors in thought. Time is a line (see above the basic compo nents of thinking); but regarding the observer there are three types of metaphors, i.e., three types of structur ing the time–observer relationship [11]. Time forms a stationary landscape in the metaphor of the orienta tion through time, the observer is also stationary. He is in the present tense, the past is behind, and the future is before him: we must look forward, but not look back; all is still ahead. Time in the timemoving metaphor passes in the observer’s direction of motion: it is time for lunch; Christmas is coming. Time in the metaphor of the observer’s direction of motion is again a landscape through which the observer moves: we passed this stage successfully. The entire history of physics can be considered as the history of exposing the intuitive ideas of the physi cal world. Only by studying physics did people learn with surprise that everyone remains at rest or in uni form and straight motion until an external force changes this state and that all bodies (heavy and light) fall with the same acceleration. Not to mention the theories of relativity and quantum physics, which even
many professionals called violence against common sense at first. An important feature of “intuitive physics” is the role of causality in the human worldview. “Recent experiments by the psychologists M. Hauser and B. Spaulding have shown that reasoning about causal powers without needing to see a long sequence of pre ceding events is by birthright the property of all pri mates, including people.” In other words, people are looking for the reasons for events and if the search for the real causes is difficult due to a lack of information, lack of knowledge, and ability to reason, they are sat isfied with simple and intelligible explanations. “The habit of creating similar forces via the imagination that influence human affairs, and forcing experience to fit them has shaped human cultures from time immemo rial, having created an extensive set of concepts, including voodoo, astrology, magic, prayer, idolatry, New Age nostrums, and other flimflams.” The Semantics of Discourse The term “discourse” is popular in recent decades, although it is often used inaccurately, but “to show erudition.” Discourse can be defined most structurally as a text or speech whose meaning cannot be ade quately understood without socialpsychological, pro fessional, and other contexts. Some linguistic charac teristics of a certain discourse (for example, a specific dictionary) are also determined by these contexts. Discourse can characterize a certain sociocultural group, political ideology, etc. Pinker does not use the term “discourse”; however, an essential part of his idea can be conveniently called discourse semantics. The basis of discourse semantics, as discussed by Pinker, is the typology of the social relationships between people that was developed by the anthropolo gist Alan Fiske [12], which consists of four types of relationships. The first type is based on solidarity, the second on power and authority, the third on social exchange, and the fourth on market pricing. The type that is based on the solidarity is public co ownership and “commonality.” It is based on the idea that others want for you the same as what you wish for yourself. What is good for one should be good for another. Commonality occurs naturally between blood relatives. In addition, common tastes or inter ests, common enemies, and communities in a foreign country can unite people in a municipal community. The mutual obligations imposed by commonality are not discussed and they are accepted by default. A good example is a marriage contract. People, as a rule, do not like to make them such that the obligations that are clearly spelled out in them can destroy the spirit of commonality in a future family. Commonality is implicit in a specific vocabulary and typical verbal clauses: brotherhood, unity, solidarity, devotion, “we are one family,” etc.
SCIENTIFIC AND TECHNICAL INFORMATION PROCESSING
Vol. 42
No. 5
2015
CONCEPTUAL SEMANTICS
The second relationship type is “ranking the authorities”; these are power, status, and domination. This ranking was fixed in the titles of the nobility in Europe, in the tables of ranks of Russia in the 18th– 19th centuries, in the corresponding strictly regulated vocabulary of addressing the nobility (“your majesty,” “your grace,” and “your high nobleness”), and in the various rituals. In our time, this ranking is preserved in the military, bureaucratic, and church hierarchies; however, rituals are preserved in the records of official receptions, the hand salute to a superior officer in the army, etc. The third relationship type is “equality matching” based on exchange and general ideas of justice. This includes agreedupon rules for the distribution of resources, the establishment of one’s turn, the exchange of services or favors, etc. Unlike the first two types, the exchange language, as a rule, is explicit, since it requires explicit agreements. Though the first three types are inherent in all human civilizations, including the most primitive, the fourth type, “market pricing,” is characteristic only for developed civilizations. It contains the compo nents of modern market economies: currency, prices, salaries, benefits, rents, rates of interest, credit, etc. Numbers, mathematical operations, digital account ing, as well as the language of official contracts, are forms of communication. Many mechanisms of the market economy seem unnatural in terms of normal cognitive processes because they do not fit the cogni tively understandable third type of “face to face exchange.” Therefore, “people believe that every object has an intrinsic fair price, that middlemen are parasites (despite the fact that they import products from distant places), and that charging interest is immoral (despite the fact that money is more valuable to people at some times than at others).” A person throughout his life, and often within 1 day, is involved in different discourses and the ability to recognize an adequate social context and corre sponding discourse is an important part of social expe rience. The pragmatics of discourse is formulated in the quotation that was given earlier in this paper: “people use language not just to transfer ideas from head to head but to negotiate the kind of relationship they wish to have with their conversational partner” [7]. The informal communication style in an inner cir cle is out of place in a formal setting. A person who is used to communicating among officials with clear ranking of authority does not understand how he must behave in the much freer academic environment, while for a former officer even the atmosphere of civil officials seems too free. The imposition of necessary discourse and appro priate vocabulary is an important component of differ ent types of mind control in advertising and political campaigns, as well as in information wars. The success of such campaigns may be explained by the fact that
311
the vocabulary of aggressive discourse, as a rule, has a strongly evaluative character (substantially different assessments are connected to various understandings of the same event or person: a revolution or a mutiny, a scout or a spy, etc., see the concept of framing above) and automatically forms the necessary relationship of accepted discourse on events, views, etc. CONCLUSIONS Pinker’s book is much richer than the theses that are described here. In addition to the basic provisions of conceptual semantics it contains a huge quantity of linguistic material and detailed discussion of many other concepts. Therefore, it can be considered as some intermediate total of more than 30 years of research on cognitive linguistics. However, conceptual semantics is of the main interest for artificial intelli gence, as described in this book. Its basic provisions are as follows: ⎯Our cognitive models are due to human nature. Each of them serves human goals; ⎯The language of thought is primary in relation to language, i.e., to verbal language; ⎯The basic components of the language of thought are a set of abstract conceptual structures that organize our experience, space, time, substance, cau sation, force, and number; ⎯The language structures the world. Our cognitive models allow us to distinguish the world as it is impor tant for the realization of the human goals in nature and society. Thought and language impose structure on the world, resulting in an intuitive worldview, both physical, and social, that can be very different from the scientific view. On the one hand, language studies allow one to reveal the features of cognitive models in everyday consciousness (thus, the subtitle of [7] “Lan guage as a window into human nature”). On the other hand, the limitation of intuitive models that can dis tort an objective worldview shows the importance of education, which allows us to compensate for this lim itation; ⎯Our speech has not only linguistic, but also social characteristics, which form a certain discourse. The correct perception of discourse leads to adequate social behavior. On the other hand, the aggressive imposition of a certain discourse is a very widespread means of mind control and social demagogy. We note, in particular, that the discourse of totalitarian ideolo gies usually resorts to the style of a statement of com monality, which is identified with agreement of opin ions and cannot occur in developed societies. The active use of such phrases as “fifth column,” “enemy of the people,” and “betrayal of the Motherland” are oriented to the condemnation of those who do not wish to share this style of pseudocommunity. Lakoff’s and Pinker’s concepts have a main issue, the original assumption that the cognitive structures
SCIENTIFIC AND TECHNICAL INFORMATION PROCESSING
Vol. 42
No. 5
2015
312
KUZNETSOV
and processes of a person are substantially determined by his physical nature. Both books are based on exten sive empirical material. Nevertheless, according to their methodological structures they differ signifi cantly. Broadly speaking, Lakoff’s book is rather deductive in its theoretical part, while Pinker’s book is inductive. Lakoff’s concept has the form of a frame work of the constructive theory of the “upper” (i.e., not coming down to neural networks) level of cognitive processes in the mind that could explain many infor mation processes of the brain, which do not fit into the computer paradigm. Pinker’s concept, which was published 20 years later, is not so systematic, but sup plements Lakoff’s harmonious concept well. Pinker, in all appearances, agrees with the basic provisions of the cognitive semantics of Lakoff. However, he criti cizes interpretations of cognitive semantics that are too straightforward, which were introduced in Lakoff’s later works that relate to epistomology and social behavior. The logic of these interpretations recognizes that “…because thinking is rooted in bodily experi ence, the concept of objective or absolute truth must be rejected. There are only competing metaphors, which are more or less apt for the purposes of the peo ple who live by them.” Pinker considers, in our opin ion rightly, that “our best science and mathematics can predict how the world will behave in ways that would be a staggering coincidence if the theories did not characterize reality.” Pinker’s thesis about the existence of the language of thought and its primacy in relationship to verbal language has to significantly influence our under standing of how information is provided in the mind and how the verbalization of our thoughts occurs. Therefore, it is very important for cognitive sciences and, in particular, for artificial intelligence.
REFERENCES 1. Church, A., Introduction to Mathematical Logic, Princ eton University Press, 1996. 2. Rakhilina, E.V., Cognitive semantics: History. People. Ideas. Results, Semiotika i Informatika, 1998, no. 36, pp. 274–323. 3. Kuznetsov, O.P., Cognitive semantics and artificial intelligence, Iskusstv. Intell. Prinyat. Reshen., 2012, no. 4, pp. 32–42. 4. Sowa, J.F., Conceptual Structures: Information Process ing in Mind and Machine, Boston, MA: AddisonWes ley Longman Publishing Co., Inc., 1984. 5. Lakoff, J., Women, Fire, and Dangerous Things: What Categories Reveal About the Mind, University of Chi cago Press, 1987. 6. Rosch, E., Cognitive representations of semantic cate gories, J. Exp. Psychol., 1975, vol. 104, pp. 192–233. 7. Pinker, S., The Stuff of Thought, Penguin Group (Viking Press), 2007. 8. Chernigovskaya, T., Cheshirskaya ulybka kota Shredingera: yazyk i soznanie (The Cheshire Smile of the Schrödinger’s Cat: Language and Consciousness), Moscow: Yazyki slavyanskoi kul’tury, 2013. 9. Wertheimer, M., Productive thinking, New York, NY: Harper. 1945. 10. Hadamard, J., The Mathematician’s Mind: The Psy chology of Invention in the Mathematical Field, Prince ton University Press, 1996. 11. Boroditsky, L., Metaphoric structuring: Understanding time through spatial metaphors, Cognition, 2000, vol. 75, no. 1, pp. 1–28. 12. Fiske, A.P., The four elementary forms of sociality: Framework for a unified theory of social relations, Psy chol. Rev., 1992, vol. 99, pp. 689–723.
SCIENTIFIC AND TECHNICAL INFORMATION PROCESSING
Translated by A. Sapronova
Vol. 42
No. 5
2015
