At Victoria Station, on a visit to London, you walk up to a (human) assistant from the Tourist Board and ask the question, ``Can you tell me how to get to the gallery in the square containing the monument?'' to which the assistant replies, ``Travelling by Underground, take the Victoria Line to Green Park, then change and take the Jubilee Line to Charing Cross.'' This apparently simple exchange involves a great deal of sophisticated cognitive processing, as we shall show in the pages that follow.
For example, the question, taken alone, appears to need a simple yes or no answer, in much the same way as do questions like ``Can you speak Gujarati?'' or ``Can you touch your toes?'' The assistant nevertheless recognizes this as an indirect request for information and responds accordingly; to have just replied ``Yes'' would have been uncooperative. Your question has not specified which gallery, square, and monument you want to visit. But you have used the definite article `the' in referring to each, so the assistant takes this to mean that you have a particular gallery, square, and monument in mind, even if you are unable to name them. The assistant draws on her knowledge of London to indentify them, and work out the best route. In fact, she has drawn on even more sophisticated knowledge than this in order to answer you, since the form of your question was ambiguous. In terms of its structure alone the question does not allow the hearer to decide whether the monument is in the gallery or in the square; that is, whether the question is more like sentence A below or sentence B:
``Can you tell me how to get to 12345678 ¯ A \> ...the gallery in the [square containing the monument]?''
compare: \> ...the corkscrew in the [drawer containing the cutlery]?''
B \> ...the gallery [in the square] containing the monument?''
compare: \> ...the compound [in the zoo] containing the wildebeest?''
The assistant must have sophisticated knowledge of sentence structure in English, since the fact that there are alternative interpretations is not formally signalled in the sentence itself. She must also know something about plausible states of affairs -- in this case that monuments are more likely to be found in squares than in galleries -- in order to determine which of the alternative interpretations is the correct one. Finally, your question has made no mention of means of transport, nor that you would prefer to know the most direct route. But the assistant knows about the best way to travel across London and understands the phrase `get to' as implicitly asking for the most direct journey. She therefore suggests you take the Underground and details the shortest way (out of probably thousands of ways) to get there.
Perhaps this example has helped you to appreciate that the ability to use language is one of the most important cognitive skills that we, as human beings, have. It is so complex and important that, as we mentioned in chapter 1, the philosopher Descartes considered it to be the primary faculty that distinguishes human beings from the lower animals. More interesting for our purposes is Descartes' further claim that language distinguishes humans from machines.
Part of the `remarkable fact' of human language that Descartes highlighted in his writings is that, unlike the fixed patterns of bird calls and bee dance, or the pure mimicry of parrots, language is infinitely productive: human beings are, in principle, able to produce and understand an infinite number of novel sentences, the only limitations in practice being that speakers and hearers eventually get physically tired and ultimately die. But those limitations have nothing to do with the facts of language. Another part of the miracle is that, again unlike parrots, we can use language to convey meanings and thoughts to others, to get at the content behind the purely physical signals of sound waves or marks on paper. Finally, humans do not merely form meaningful sentences, but can give an `appropriately meaningful answer to what is said'; that is, they can use language in ways appropriate to the context. Linguists usually talk about these three facets of language under the headings syntax, semantics, and pragmatics, respectively. We shall follow these divisions in sections 5.2-5.4 below.
It is largely from Descartes and the seventeenth century Port-Royal school of French philosophers that Noam Chomsky, by far the most influential linguist and language theoretician of the twentieth century, drew the fundamental notions that led to the establishment of linguistics as a cognitive science, or more specifically, in Chomsky's words, as ``a branch of cognitive psychology.'' Descartes was a rationalist; he held that human beings are endowed with `innate ideas' that determine and constrain the form of human knowledge in quite specific ways, and that it is these innate ideas that distinguish the creative intelligence of humans from the mechanical behaviour of animals and machines. Those innate ideas associated specifically with language Chomsky calls `universal grammar', ``the system of principles, conditions, and rules that are elements or properties of all human languages not merely by accident but by ... biological ... necessity'' (Chomsky, 1975, p. 29). It is this `universal grammar' that accounts for the invariance of certain properties -- `linguistic universals' -- across all languages, and for the remarkable speed with which young children acquire their own native tongue. Children learn their first language so quickly, Chomsky argues, because they in some sense already `know' what human languages look like: they are biologically equipped with what he calls a `Language Acquisition Device'. If such principles are indeed innate, then a compelling reason for studying human language is that it may gives us insights into the structure of the mind itself.[+]
Chomsky has drawn a distinction between linguistic competence and linguistic performance, the first being an ideal speaker-hearer's knowledge of her language, and the latter a person's actual use of language in real situations. The primary task of linguistics, in Chomsky's view, is to characterize, in the form of a grammar of the language, the ideal speaker-hearer's intrinsic competence. The emphasis in artificial intelligence approaches to natural language has been slightly different. Of course it is important to describe the formal properties of languages; but such descriptions do not, in themselves, say how you would go about producing an actual utterance, or understanding the utterances of others. Since the main thrust of natural language processing in artificial intelligence has been to design machines that can produce and understand human languages, there has been a far greater concern among cognitive scientists to give accounts of how we put to use the internalized knowledge we have of our mother tongues. This division is mirrored in the distinction cognitive scientists make between grammars and parsers, which are programs that make use of grammatical knowledge. In most computational natural language systems, the grammar of the language and the parser that works on it are written as separate and independent components -- the grammar is encoded as a declarative data structure, the parser as a procedural program. For reasons of simplicity, however, we do not make this separation in the simple system that we develop in the appendix to this chapter.