Abstract
This chapter describes the functioning of natural language. To this purpose, the robot Curious – introduced in Chap. 3 as an artificial cognitive agent without language – is equipped with additional components needed for natural language communication between the robot and its wardens.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Traditional grammar and theoretical linguistics (1.2.1) avoid the modeling of cognition. Their analyses are purely sign-oriented and concentrate on the structural properties of expressions such as word forms, sentences, or texts. For this, dictionaries have been compiled listing the words of a language; generative syntax grammars have been developed which try to formally distinguish between well-formed and ill-formed sentences; and the meanings of sentences have been characterized in logical semantics as relations between expressions and the world – omitting the cognitive structure of the speaker-hearer. When the goal is to model successful language use, however, an agent-oriented approach cannot be avoided. It must provide a computational reconstruction of the cognitive processing in the speak and hear modes which is functionally equivalent to the natural prototype.
- 2.
- 3.
For the purpose of programming, this solution has turned out to be insufficient (CLaTR’11, Chap. 4).
- 4.
Without detracting from their merits in other areas, such a trivialized treatment of reference may be found in SHRDLU (Winograd 1972) and in Montague grammar (Montague 1974) . Winograd treats an expression like blue pyramid by ‘gluing’ it once and for all to a corresponding construct in the toy world of the SHRDLU program. Montague defines the denotation of a predicate like sleeps in the metalanguage via the denotation function F for all possible worlds and moments of time. In either case, no distinction is made between the meanings of language expressions and corresponding sets of objects in the world.
Binding language expressions to their referents in terms of definitions (in either a logical or a programming language) has the short term advantage of (i) avoiding a semantic analysis of language meaning and (ii) treating reference as an external connection – like the dotted line in 4.2.1. The cost for this is high indeed: such systems are in principle limited to being closed. See Chaps. 19–22, especially Sect. 20.3, for a detailed discussion.
- 5.
Translated from German unmittelbar (immediate) and mittelbar (mediated). Wahrig (1986/1989) defines unmittelbar as ‘ohne örtl. od. zeitl. Zwischenraum’ (without spatial or temporal distance).
- 6.
Immediate reference may occur outside the communication prototype 3.1.2, for example, when a hearer finds a note on her desk, saying: Have you found the cookies in the right-hand drawer?
- 7.
For example, when speaker and hearer talk about the person of J.S. Bach (1685–1750), they refer to a contextual structure for which there is no counterpart in the current real world. Another form of mediated reference is Curious’ reference to objects in a state which was current in the past, as in How many red triangles did you find yesterday?
- 8.
These concerns underlie the laborious arguments guarding against possible accusations of ‘psychologism’ in the writings of Frege, among others.
- 9.
One might argue that Wittgenstein concentrated on the first notion in his early (1921) work and on the second notion in his late (1953) philosophy. Rather than functionally integrating expression meaning into utterance meaning, as in PoP-1 (4.3.3), Wittgenstein opted to abandon his first approach. See also the discussion of ordinary language philosophy as exemplified by Grice in Sect. 4.5 and the discussion of semantic ontologies in Chap. 20.
- 10.
- 11.
For example, the meaning1 of the word square defined in 4.2.2 is a concept type which exists independently of any possible referents, in either the agent-internal context or the agent-external task environment. Correspondingly, the square objects in the world and their reflexes in the cognitive agents’ context do not depend on the existence of a word with a corresponding meaning1 – as demonstrated by the nonlanguage version of Curious in Chap. 3.
This independence between concept types and corresponding concept tokens holds only for the secondary use of concept types as language meanings1 which are lexically bound (by convention) to the surface types of symbols, as shown in 4.2.2. In their primary function as contextual types for certain parameter constellations, in contrast, concept types are the precondition for the derivation of concept tokens (3.3.5).
- 12.
- 13.
See Sect. 6.5 for a distinction between syntactic, semantic, and pragmatic ambiguities.
- 14.
- 15.
The SC-I principle was first described in Hausser (1978). As shown in SCG, it may be interpreted formally as restricting Montague’s homomorphism condition regarding the relation between a categorial syntax and a formal semantics based on a typed lambda calculus. The relation between the SC-I principle and the mathematical notion of a homomorphism is described in Sect. 21.3, in which the informal, intuitive version of 4.5.1 is supplemented by the formal variant SC-II (21.3.6).
- 16.
A description of the post-transformational systems of GB, LFG, and GPSG may be found in Sells (1985) . Yet another variant of nativism is HPSG as the continuation of GPSG.
- 17.
Some followers of Chomsky have recognized this as a weakness. However, their belated attempts to provide transformations with a genuine functional role have not been successful. Chomsky has always rejected such attempts as inappropriate for his nativist program (e.g., Chomsky 1965, p. 9).
- 18.
Transformations were later replaced by similar mechanisms. When asked in 1994 about the frequent, seemingly radical, changes in his theories, Chomsky pointed out that the ‘leading ideas’ had never changed (personal communication by Prof. Dong-Whee Yang, Seoul, Korea 1995). See also Chomsky (1981), p. 3, in the same vein.
- 19.
The surface compositional treatment is based on separate derivations of the active and the passive. If they happen to be paraphrases, this may be expressed by establishing semantic equivalence on the level of meanings1 (4.4.6).
- 20.
Bresnan (ed.) (1982).
- 21.
Gazdar et al. (1985).
- 22.
- 23.
According to Levinson (1983), pp. 227f.
there are strong parallels between the later Wittgenstein’s emphasis on language usage and language-games and Austin’s (1962:147) insistence that “the total speech act in the total speech situation is the only actual phenomenon which, in the last resort, we are engaged in elucidating”. Nevertheless Austin appears to have been largely unaware of, and probably quite uninfluenced by, Wittgenstein’s later work, and we may treat Austin’s theory as autonomous.
That Austin, who was 22 years younger than Wittgenstein, was “largely unaware” of Wittgenstein’s writings, is attributed by Levinson to their teaching at different universities: Austin was at Oxford and Wittgenstein 66 miles away at Cambridge. The train once connecting the towns via Bletchley was known as the “brain line.”
- 24.
- 25.
This reduction is epitomized by ordinary language philosophy’s imprecise and misleading formula
Meaning is use.
The Slim theory of language is also based on use, but on the use of literal expression meanings1 relative to a context of interpretation, resulting in the speaker’s utterance meaning2 (PoP-1, 4.3.3).
- 26.
See for example Searle (1969) , pp. 44f.
- 27.
See also Austin (1962) , pp. 121f.
- 28.
Rather than applying the type/token distinction to conventions, the Slim theory of language applies it to recognition and action. In contextual recognition, for example, the types arise as classes of similar parameter constellations. Once a type has evolved in this way it is used to classify corresponding constellations of parameter values, resulting in tokens instantiating the type.
In the extension to language, the types are used in a secondary function, namely as the meaning1 of symbols. The interpretation of symbols is based on matching these meanings1 (concept types) with contextual referent structures (concept tokens).
The principle of internal matching between concept types and concept tokens aims from the outset at handling the spontaneous use of language to express new meanings2 relative to new contexts. Conventions are used only for fixing the relation between the language surfaces and their meaning1 inside the speaker-hearer (in agreement with de Saussure’s first law, 6.2.2).
This straightforward explanation of the primary origin and function of types and tokens on the contextual level and the secondary functioning of the types as the word meanings1 of language surfaces cannot be transferred to Grice’s speech act theory. The reason is that the speaker’s intentions are not accessible to the hearer’s recognition as characteristic parameter constellations – in contradistinction to the Slim-theoretic meaning2 derivation based on lexical types and contextual tokens.
- 29.
Regarding its mathematical properties, a formal definition of Grice’s approach has yet to be provided. What such a formalization could look like is indicated by another system of ordinary language philosophy and speech act theory, namely that of Searle and Vanderveken (1985) . Depending on one’s approach, this formalization either may be shown to be too imprecise for drawing any firm mathematical conclusions, or some particular aspect of it may be isolated for demonstrating high complexity.
References
Austin, J.L. (1962) How to Do Things with Words, Oxford: Clarendon Press
Barcan-Marcus, R. (1960) “Extensionality,” Mind 69:55–62, reprinted in L. Linsky (ed.), 1971
Bresnan, J. (ed.) (1982) The Mental Representation of Grammatical Relation, Cambridge: MIT Press
Chomsky, N. (1965) Aspects of the Theory of Syntax, Cambridge: MIT Press
Chomsky, N. (1981) Lectures on Government and Binding, Dordrecht: Foris
Gazdar, G., E. Klein, G. Pullum, and I. Sag (1985) Generalized Phrase Structure Grammar, Cambridge/Oxford: Harvard University Press/Blackwell
Grice, P. (1957) “Meaning,” Philosophical Review 66:377–388
Grice, P. (1965) “Utterer’s Meaning, Sentence Meaning, and Word Meaning,” Foundations of Language 4:1–18
Hausser, R. (1978) “Surface Compositionality and the Semantics of Mood,” reprinted in J. Searle, F. Kiefer, and M. Bierwisch (eds.), 1980
Hausser, R. (1980) “The Place of Pragmatics in Model-Theory,” in J.A.G. Groenendijk et al. (eds.)
Katz, J., and P. Postal (1964) An Integrated Theory of Linguistic Descriptions, Cambridge: MIT Press
Levinson, S.C. (1983) Pragmatics, Cambridge: Cambridge University Press
Montague, R. (1974) Formal Philosophy, New Haven: Yale University Press
Pollard, C., and I. Sag (1987) Information-Based Syntax and Semantics Vol. I, Fundamentals, Stanford University: CSLI Lecture Notes, Vol. 13
Pollard, C., and I. Sag (1994) Head-Driven Phrase Structure Grammar, Stanford/Chicago: CSLI Stanford/The University of Chicago Press
Searle, J.R. (1969) Speech Acts, Cambridge: Cambridge University Press
Searle, J.R., and D. Vanderveken (1985) Foundations of Illocutionary Logic, Cambridge: Cambridge University Press
Sells, P. (1985) Lectures on Contemporary Syntactic Theory, Stanford University: CSLI Lecture Notes, Vol. 3
Wahrig, G. (1986/1989) Deutsches Wörterbuch, Munich: Mosaik Verlag
Winograd, T. (1972) Understanding Natural Language, San Diego: Academic Press, Harcourt Brace Jovanovich
Wittgenstein, L. (1921) “Logisch-philosophische Abhandlung,” Annalen der Naturphilosophie 14:185–262
Wittgenstein, L. (1953) Philosophical Investigations, Oxford: Blackwell
Author information
Authors and Affiliations
Exercises
Exercises
Section 4.1
-
1.
What components are required to extend the nonlanguage version of Curious presented in Chap. 3 into a robot communicating in natural language?
-
2.
In what sense are the components of language processing in Curious a specialization of its contextual cognitive components?
-
3.
What is the relation between natural language defined as a set of grammatically analyzed expressions, and the cognitive structure of the speaker-hearer using the language? Explain your answer in three to four pages.
-
4.
Why does the construction of artificial cognitive agents have a methodological impact on the formation of linguistic theory?
Section 4.2
-
1.
What is the internal aspect of reference?
-
2.
Explain the cognitive [2+1] level structure of reference in natural language.
-
3.
On what basis does the hearer establish reference if the speaker uses an expression not heard before (e.g., Mars mushroom) to refer to an object not seen before?
-
4.
Why is the handling of reference nontrivial in the case of Curious, but trivial in the case of SHRDLU? How does this difference depend on the distinction or nondistinction between task environment and problem space?
-
5.
In what respect can SHRDLU do more than Curious in its present form? What would be required to combine the different merits of the two systems?
Section 4.3
-
1.
Explain which letters in the acronym of the Slim theory of language relate directly or indirectly to the [2+1] level structure of reference?
-
2.
What is the difference between immediate and mediated reference?
-
3.
Describe the connection between reference and cognitive processing.
-
4.
What is the difference between the speak mode and the hear mode in natural language communication?
-
5.
Describe the connection between the literal meaning1 of a language expression and the speaker meaning2 of an utterance.
-
6.
Given that the time linearity of natural language signs is represented from left to right, can you explain why the hearer is placed to the left of the speaker in 4.3.2 and 5.2.1?
Section 4.4
-
1.
Who was Frege and when did he live? Explain the principle that carries his name.
-
2.
Does Frege’s principle relate to the speaker’s utterance meaning2 or the expression’s literal meaning1?
-
3.
Why are ambiguity and paraphrase apparent exceptions to Frege’s principle? Which properties of the analysis make it possible to eliminate these exceptions? Explain your answer using concrete examples of ambiguity and paraphrase.
-
4.
Name other areas in which confusing identity and equivalence has led to problems (cf. Barcan-Marcus 1960).
-
5.
What is the relation between the principle of surface compositionality and Frege’s principle?
Section 4.5
-
1.
Give nine different examples of transformations and show how they violate surface compositionality.
-
2.
Why can Frege’s principle be applied only to the deep structures of transformational grammar, but not to the surface structures?
-
3.
What is a methodological objection to applying Frege’s principle to deep structures?
-
4.
Describe the definition of meaning by Grice and explain why it is not suitable for computational linguistics.
-
5.
Compare the analysis of meaning in the theory of Grice and the Slim theory of language. Explain the different uses of the type/token distinction and the different definitions of sentence and utterance meaning in the two theories.
-
6.
What can a concrete implementation of Curious as a talking robot do for an improved understanding of natural language communication?
-
7.
Explain the criteria for successful human-computer communication in natural language.
-
8.
Compare the language processing of Eliza, SHRDLU and Curious.
Rights and permissions
Copyright information
© 2014 Springer-Verlag Berlin Heidelberg
About this chapter
Cite this chapter
Hausser, R. (2014). Language Communication. In: Foundations of Computational Linguistics. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-41431-2_4
Download citation
DOI: https://doi.org/10.1007/978-3-642-41431-2_4
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-642-41430-5
Online ISBN: 978-3-642-41431-2
eBook Packages: Computer ScienceComputer Science (R0)