Peter Svenonius

Chomsky: Language is not designed for communication

Chomsky has suggested that language is not designed for communication (e.g. Chomsky 1966, more recently in Chomsky 2010). Recently, Piantadosi, Tily, and Gibson (2012) attempt to show that Chomsky is wrong about this. They suggest that their findings provide “theoretical and empirical arguments against recent suggestions that core features of linguistic systems are not designed for communication” (p. 280). Their argument, however, is seriously flawed, and they don’t seem to have understood Chomsky’s point.

Chomsky studies language from a biological perspective. Thus, for Chomsky, language is the human capacity for the kind of digitized, structured symbolic computations that we perform every time we construct a sentence. So when he says that it is not designed for communication, he is talking about evolutionary design. The question is, did the human capacity for language evolve under selective pressure for better communication among individuals in the species, or did it evolve in some other way? Chomsky suggests that the evidence, though scant, points away from communication as a driving force in the evolution of this particular talent that all and only humans have, of the animals on this planet.

The Silence of the Gaps

Chomsky mentions several examples of fundamental features of language that you wouldn’t wish for in a well-designed communication system. As one example, he mentions the silence of gaps, positions in which a moved element is interpreted but not pronounced. For example, in the most salient reading of (1), the gap is after ‘wanted’:

(1)      What did John say that Mary wanted?

But in (2), the most natural interpretation includes a gap after ‘say’:

(2)      What did John say that offended Mary?

The silence of gaps is a systematic feature of language but is known to lead to parsing problems. A simple design feature would be to include pronounced markers at the point of the gaps, so that they are unambiguous, as in (1′) and (2′) (these are not natural English but English-like sentences and I signal this by enclosing them in quotes):

(1′)     ‘What did John say that Mary wanted “it”?’

(2′)    ‘What did John say “it” such that “that” offended Mary?’

Ambiguity

The silence of gaps even leads to ambiguities which must be resolved by context, for example in (3).

(3)      What did John say that Mary believed?

The sentence in (3) could either mean (3′) or (3”):

(3′)     ‘Of what did John say this: that Mary believed “it”?’

(3”)    ‘What did John say “it” such that Mary believed “that”?’

Scope Ambiguity

Chomsky also mentions scope reconstruction, for example in (4):

(4)      Everybody didn’t like the movie

In (4), the meaning can either be as in (4′) or as in (4”):

(4′)     Everybody disliked the movie

(4”)    Not everybody liked the movie

Unambiguous examples like (4′) and (4”) are not hard to come up with, but language systematically causes us to produce examples like (4). In context, we don’t usually notice, but sometimes such sentences lead to confusion. If language were designed for communication, it would seem easy to include a basic design feature to avoid these ambiguities, especially given that the unambiguous alternatives are so easy to produce.

I have used a systematic scope ambiguity in the title of this essay. It is repeated in (5):

(5)      Language is only used for communication.

This can be understood either as (5′) or as (5”):

(5′)     Language is used only for communication.

(5”)    Language is used for communication, but doesn’t have any other special relationship to communication.

This latter reading is salient in pronunciation if you stress *used* in (5). The scope of only in writing is systematically ambiguous in this way, and in some circumstances it is ambiguous in pronunciation as well. All of these ambiguities could easily be avoided if only were obligatorily pronounced next to the phrase it is focusing. Then (5′) would only have the meaning it has, and (5) would only have the meaning of (5”)—oops, I should have written, “(5′) would have only the meaning it has, and (5) would have only the meaning of (5”)” (to see the unwanted readings of the sentences as written first, stress have).

For Chomsky, the fact that language is designed in a way that leads to systematic structural ambiguities is a clue that competitive pressures did not act on alternative genetically based communication systems over an evolutionary timescale to result in the emergence of human language. Instead, evolution may have favored a digitized symbolic computational system for thought, with communication as a subsequent application of that ability. The ambiguities noted above do not occur to the person having the original thought; it is only in the translation of the thought to something pronounceable that the ambiguity arises, for the listener.

Ambiguity as a design feature

Piantadosi, Tily, and Gibson take up the issue of ambiguity and think that they have found an argument against Chomsky. They react to his claim that ambiguity is a sign of poor design and suggest that Chomsky’s “perspective on ambiguity is exactly backwards” (p. 281).

Their first argument is that since informative context always reduces ambiguity, it is inefficient to redundantly code information into a message that could be extracted from context: “the most efficient communication system will not convey information already provided by the context” (p. 282).

So, they conclude, rather than being a design flaw, “ambiguity is a core functional component of language” (p. 284). They seem here not to be distinguishing ambiguity from vagueness (the distinction is fairly standard in linguistics, see e.g. Binnick 1970 for early discussion, Zhang 1998 for a more recent treatment). Clearly, a communication system would be inefficient if it had to include full specifications of information which is otherwise deducible. In fact, it would be impossible to encode all of the meaning which is intended in a simple utterance. But this doesn’t suggest that building in structural ambiguities of the type discussed above should be a positive design feature for language—there doesn’t seem to be much communicative efficiency gained in making gaps silent, or in making quantifiers (like everybody) and focus operators (like only) ambiguous in scope.

Short words

Their second argument is primarily based on the following observation: short ‘easy’ words are more likely to be ambiguous than long ‘hard’ words. To illustrate, imagine that you come across the following two words in an unknown language:

(6)      ta

(7)      tankeeksperiment

Guess which word is more likely to have several different senses. Did you guess (6)? If not, then Piantadosi, Tily and Gibson’s discovery may come to you as a surprise. The discovery that they build their argument on is that a word like (6) (which means ‘take’ in Norwegian) is more likely to have several meanings than a word like (7) (which means ‘thought experiment’). They show that this is statistically true whether the words are compared in terms of length ((6) being shorter than (7)), phonotactics ((6) is easier to pronounce than (7)), or frequency ((6) is more frequent than (7)). They suggest, in their conclusion (p. 290): “These results are hard to explain with anything other than a theory based on efficient communication: what theory would posit that ambiguity should preferentially be found in these linguistic units, but not that it results from pressure for efficiency?”

It seems that they have entirely missed Chomsky’s point about ambiguity. The fact that short, easy to pronounce forms are more likely to be adopted as word forms is already described by Zipf’s (1949) principle of least effort. The same principle ensures that a host conventionally offers a guest a choice of a couple of salient beverage alternatives appropriate to the time of day and occasion, rather than to list every beverage in the house or available at nearby stores.

Suffixes

A simple confound that Piantadosi, Tily, and Gibson don’t appear to have controlled for is morphological complexity. Simple stems like judge in English tend to be usable as nouns and as verbs. That counts as ambiguity by their reckoning. Words with category-specific suffixes, like evaluate and evaluation, tend not to have the flexibility of simple stems (evaluate is a verb, and evaluation is a noun). The suffixes signal the category, and also make the words longer and more complex to pronounce.

Even affixes which do not derive category are often category-specific, so they will restrict the meaning of the whole; so prejudge can only be a verb. Also, compounds, which the paper apparently counted, will tend to add specific dimensions of meaning in ways that will restrict ambiguity; so bench-press doesn’t have as many senses as press does.

But even if you controlled for these facts, there is presumably a tendency for short words to have multiple meanings more often than long words. The question is in what sense this is a matter of design, rather than lack of design.

Homophony and Polysemy

Recall that in Chomsky’s sense, language is the biological capacity in humans for constructing structured symbolic representations of thought. If that system has no principle against homophony, then only matters of usage will prevent homophony from arising in conventions of language use. So short, easy to pronounce words will tend to have multiple meanings because of a fact about usage of language, not because of the design of the language capacity. Piantadosi, Tily, and Gibson don’t seem to think that they are talking about the historical emergence of conventions. They write that language should be considered as “a cognitive system designed in part for communication” (p. 290). But a cognitive system suggests a biological system, and so it seems that think that they have found evidence that the biological capacity for language actually seeks ambiguity as a way of making the lexicon more efficient.

It is easy to imagine a language capacity which was designed to avoid homophony, without becoming less effective at communication. A sentence like (8) contains an ambiguity of the kind studied by Piantadosi, Tily, and Gibson because bank can mean either a financial institution or the land along the side of a river.

(8)      There’s a fire near the bank!

This ambiguity could be avoided if instead of ‘recycling’ already used forms, the language capacity preferred to modify short, easy to pronounce to produce forms which are not already in use. When the French word banque was being introduced into English, instead of adding more meanings to the English word bank, why not change it to bonk, or pank, or bant? These are just as short and easy to pronounce. Piantadosi, Tily, and Gibson only show that it would be less efficient to substitute a longer or more difficult to pronounce form. Rather than being “hard to explain,” it seems that the observed facts are what one would expect if language simply didn’t care one way or another about homophony.

The two senses of bank are semantically unrelated homophones. Piantadosi, Tily, and Gibson also counted something vastly more common, namely multiple senses enumerated in dictionary entries. A dictionary will distinguish senses of fly such as 1. ‘to engage in flight’; 2. ‘to rise or be carried through the air by the wind’; 3. ‘to move or be sent through the air with great speed’; 4. ‘to move with great speed, rush, or dart; 5. ‘to pass by swiftly’; … just mentioning the first few main entries (from the American Heritage College Dictionary, Third Edition). Again they found that short words tend to have more senses. Here it seems that communicative efficiency would be reduced if different forms were introduced, but not for any reason that Piantadosi, Tily and Gibson discuss. In this case, we are probably looking at many cases of underspecification and vagueness rather than actual ambiguity. A word like fly is consistent with volition, but doesn’t entail it, and the dictionary lists both volitional and nonvolitional cases because they are both frequent and salient, not because lexical ambiguity is involved in the human competence that underlies knowledge of the word.

In sum, it seems that what was intended as a direct frontal assault on Chomsky’s conjecture about the evolutionary origins of language misses the mark entirely.

References

Binnick, Robert I. 1970. Ambiguity and vagueness. In Papers from the Sixth Regional Meeting, Chicago Linguistic Society, 147–153. Department of Linguistics, University of Chicago.

Chomsky, Noam. 1966. Cartesian Linguistics. Harper & Row, New York.

Chomsky, Noam. 2010. Some simple evo-devo theses: How true might they be for language? In Approaches to the Evolution of Language, ed. by R. K. Larson, V. M. Deprez, and H. Yamakido. Cambridge University Press, Cambridge.

Piantadosi, Steven T., Harry Tily, and Edward Gibson. 2012. The communicative function of ambiguity in language. Cognition 122: 28–291.

Zhang, Qiao. 1998. Fuzziness – vagueness – generality – ambiguity. Journal of Pragmatics 29.1: 13–31.

Zipf, George K. 1949. Human Behaviour and the Principle of Least Effort: An Introduction to Human Ecology. Addison-Wesley, Cambridge, Ma.