UofA

Spring 2004

Ling/Phil 596D: Topics in Linguistics and Philosophy

Heidi Harley and Massimo Piattelli-Palmarini

Compositionality

Wednesday, February 4

Handout 1 (M. Piattelli-Palmarini (comments by Heidi in underlined red)

Three positions: Chomsky, Higginbotham, Fodor

1-1 Bird’s eye Chomsky

He is (legitimately, I think) considered to be a “lexicalist” (ever since, at least, Remarks on nominalization (Chomsky, 1972), but arguably from the very beginning (Chomsky, 1985) originally 1955, notably in the central role always given to selectional restrictions and sub-categorizations). Though not under the narrow characterization that this position has taken in recent years (after the “constructivist” alternative of Hale and Keyser (Hale and Keyser, 1993, 2002). We will hear a lot more from Heidi about this. The primary ingredients are items from the mental lexicon, which project structures upward, from the inside out, cyclically. It’s hard to be more “compositional” than that.

The link between syntax and semantics has been progressively displaced: D-structure, then S-structure, then LF, and now a whole “interface” imposing weird constraints (interpretable features) that the NS (Narrow Syntax) computational system has to do its best to satisfy.

Again and again (ever since (Chomsky, 1995) and in the present LSA draft) he has insisted that items in the mental lexicon have no referents. No minimally consistent material entity does meet the requirements, nay, no minimally consistent material entity could meet the requirements (older cases: city, book. More recently: river and highway). Yet, any child just loves to hear stories about donkeys being turned into rocks (but still being donkeys), then turning back into donkeys again, and other strange transmogrifications typical of fairy tales, which plausibly involve subtle criteria such as “continuity of psychological sensations” or the like. Lexical meanings are always, irreducibly, objects under a certain point of view, often under multiple co-present points of view (abstractions and material things, sets of buildings and the layer of air that hovers over them, and activities taking place there, and a style of living, etc.)

The only posit that is tenable is the internal structure of the speaker-hearer, a complex, abstractly characterizable, computational-derivational apparatus, optimal if left alone, that interfaces with other cognitive apparatuses (the articulatory-perceptual one, via PF, and the conceptual-intentional one, via LF), satisfying the constraints that they impose. Any notion of a “relationship” between the speaker-hearer’s internal systems and some abstraction (even the LOT) is un-scientific, and has to be eliminated (much to the same effect, see the recent papers by Halle and Bromberger contra types in phonology). His own older terminology “Knowledge of Language” is (as we said last week) to be taken with great precaution.

These days, Chomsky is relentless in reminding us that we should not forget the important lesson that certain analytic philosophers (Moore, Strawson and Co.) have taught us. Ordinary language bamboozles us into giving reality to mere ways of saying (knowledge of language, and representations being among them). Everything is to be cashed in terms of derivations, eliminating representations altogether, as well as labels, indices, and so on (except as innocent, easily convertible abbreviations).

Also: Strictly speaking, only people refer to this and that. The extension of “referring” to apply to items in the lexicon is an improper extension, a trick of ordinary language.

1-2 Bird’s eye Fodor

No need to go over this again. Just a reminder: It cannot just be an assumption that natural language (L) is compositional, and that there is a strictly compositional level of representation (LF) of L, such that everything is made explicit at that level. This is a very strong hypothesis, that forces one to hypothesize hugely complicated underlying structures, with a huge amount of silent components (functional categories of all sorts, projections of all sorts) just because the theory imposes that LSs are strictly compositional. A theory that imposes so many diversified and complex posits ought to reconsider the basic assumption that make these inevitable, i.e. the assumption of compositionality.

However, the LOT is compositional (this is still non-negotiable), and so is syntax (the algorithmic construction of sentence types from sentence tokens, as Jerry puts it).

NOTE: Probably Chomsky would disagree about the very existence of sentence types (other than as a manner of speaking). The derivational apparatus surely has its own modi (yes, plural) operandi, but there is no such thing as sentence “types” out there (or in here) as distinct entities that the system constructs.

1-3 Bird’s eye Higginbotham

The assigners of meaning are positions in derivations (nodes and roots, if one takes the tree representation seriously). There is (in his words) determinism: For each point in a tree, its meaning is unique (see his paper in the reserve). There is a fact of the matter as to what a linguistic expression really means. In the ideal case, every speaker of English may get it wrong (remember No eye-injury is too trivial to ignore, or the third meaning of I almost had my wallet stolen), yet it means what it means, and that can be shown to be the case via careful syntactico-semantic analysis (via some sort of mini-theorems). He insists that what is crucial is not “meanings,” whatever they may be, but rather what it is to know the meaning of an expression (his former students Larson and Segal have used this expression as the title of their excellent textbooks of semantics Knowledge of Meaning).

A crucial passage from his present paper:

“As is customary, even if surely an idealization, I will assume that knowledge of the meaning of an expression takes the form of knowledge of a certain condition on its reference (possibly given various parameters) that is uniform across speakers; that is, that we can sum up what is required to know the meaning of an expression in a single statement, and that we do not adjust that statement so as to reflect, for instance, the different demands that may be placed upon speakers, depending upon their age, profession, or duties”.

As a consequence, the immediate first-blush intuitions of the single native speaker are not always the supreme judges (at least not in every case). The very notion of “elucidations of meaning” (Higginbotham, 1989) entails a work of clarification that goes beyond direct individual intuitions. Educated (parametrized, across-speakers, context-independent) intuitions is what you need.

In an e-mail message to me (October 2002), Jim says: “Elucidations are statements of what one knows who knows the meaning of something. They are not paraphrases of anything. They do, however, play the role of one notion of senses as in Frege, namely that of "cognitive

significance." This is close to one version of "mode of presentation”."

Jim has insisted, contra Fodor, that, although definitions (or paraphrases) do not exhaust lexical meanings, yet, this does not entail that they are worthless. They may well contribute to the elucidation of meaning, and be a component of meaning.

In that e-mail, Jim also says:

“Fodorian atomism is consistent, I think, with there being elaborate things one has to know to know the meaning of something, even if there are no "definitions," in some strict sense.” I agree

An example of elucidation from Jim: The meaning of “heed” (I had wondered about this meaning, untranslatable into Italian with a single word):

The data:

heed a warning/command

heed advice

Heed my words!

heed the man

*heed an order not so sure that this is * for me...

heed the instructions

*heed the book

heed the Bible

heeded naought/*nothing again, not sure about this -- 'he heeded nothing and no one' not so bad...

*heed the stove

heed the advanced passed pawn

Jim adds: I think (that is to say, it accords with my judgement: I haven't looked up the word). The curiosity is that, while 'heed' does indeed mean to pay attention to, or to take into

consideration, it applies only to acts of speech or to portents (direct, like

warnings, or from the source, like the Bible: hence the difference between 'the

book' and 'the Bible' as objects, and hence, while one can pay attention to the

stove, or take it (its condition) into consideration, one can't heed the stove

--- naturally, all the starred examples can be rescued by a sufficiently thick

context fair enough); and the speech acts must be acts of advisement or exhortation (hence

you can heed a command, but not an order, for an order must be obeyed hmm -- very subtle. I don't find that an order must be obeyed so much as all that). Your

can heed a threat (the advanced passed pawn), and in this sense you could heed

the weather, I suppose; but you can't heed an ordinary object. Ok, this accords fairly well with my own intuitions about 'heed'; differences above are not with 'heed' but with ease of construing the relevant contexts, restrictions on 'order' etc. I am ready to

believe, upon reflection, that the word is not translatable!

He tends to minimize Chomsky’s challenge of the existence of referents. The paradoxes are explicable via “a change in context”. Within the same sentence, context can change, (I MPP add that usually the standard cases are, in fact, generated via adjunction or coordination). So, you have book as an abstract and then as a concrete, London as a set of buildings, then as a way of life etc. One might say that a similar phenomenon is at work in the interpretation of nominal compounds, where the meanings are clearly compositional in some sense, but highly context-dependent; the usual example of alligator shoes might be relevant. Both concepts are involved in whatever the final meaning is, and each contributes their own atomistic meaning (hence the whole thing is compositional) — but the relationship between the two is entirely up for grabs (well, not entirely; perhaps constrained by some knowledge like Pustejovsky's qualia structure, though of course that is encyclopedic, not lexical, knowledge).

Fodor also minimizes the challenge, on pragmatic grounds: If we say we meet in London (UK) next Tuesday, then we damn meet THERE next Tuesday. Where is the problem? Fodor said to me in an aside during his visit of a couple of years ago that he believed in an extensionalist semantics for language; it's just that the extensions themselves are in the head -- they're the entities etc. in one's mental model of the world. Words do directly refer to their extensions in-the-head. And of course an in-the-head London can have all the relevant properties. I think I tend to agree with that position.

2-1 Higginbotham and compositionality in general

The risk of the standard compositionality thesis is that “it verges on the trivial”.

“For, there being nothing else but the parts of a sentence and the way they are combined to give the interpretation of a sentence, what else could interpretation be determined by? The thesis is not quite trivial, however, inasmuch as it presupposes that there is such a thing as “the interpretation of a sentence or discourse” in a proper sense.” (p1 of the paper in the reserve, emphasis original)

There being a “proper sense”, and the importance of this fact, is central to his semantic theory.

He has insisted, over the years that what is in the lexicon is a highly theoretical issue. One is not entitled (contra Grimshaw, and siding with Larson, for instance) to discharge a huge portion of syntax onto the lexicon, then (of course) claim one has reduced syntax, somehow. Abso-bleedin'-lutely.

In particular, when and why two lexical items are, or are not, to be considered as synonymous is a serious matter. One of his examples is autobiography versus the history of the life of its own author.

There is a split here between syntax and semantics.

John wrote his autobiography

*John wrote his history of the life of his own author.

The second expression is:

(15) (the x) history of the life of its own author(x) & R(John,x)

where R can be anything,

whereas John’s autobiography is:

(16) (the x) autobiography(-of) (John,x)

autobiography has two open positions, history of the life of its own author only one. At

the same time, these expressions are synonymous, considered as predicate nominals.

Syntactic headedness and semantic headedness usually map consistently and biunivocally one onto the other (this is the default hypothesis that the child brings onto her learning of the local language), but there are interesting exceptions. Alleged thief is syntactically a modification of the noun, but it is semantically a modification of the adjectival phrase. Other examples in the paper.

The lesson: No need to introduce higher semantic types. Switched-headedness between syntax and semantics accounts for all these cases. Syntactic headedness has to do with licensing the presence of the other. Semantic headedness has to do with argument-taking. Whichever takes the other as argument is the head. Usually one head maps onto the other head, but not always. A nominal projection alleged thief, can have the noun as, in fact, the complement of the adjective. Semantically this is very clear: An alleged thief is a person x of whom it is alleged: that x is a thief. Other examples (with adverbials such as eagerly) make all the standard inferences come out right.

2-2 Higginbotham on syntax and semantics

A famous thesis of his (ever since On semantics 1985 (Higginbotham, 1987; Higginbotham, 1985) is that semantics is partially insensitive to questions of grammaticality. Likely and probable are synonymous, though probable does not admit subject-to-subject raising. This explains the quartet (3)-(6) (taken from Gazdar (1985)):

(3) It is likely that Alex will leave.

(4) It is probable that Alex will leave.

(5) Alex is likely to leave.

(6) *Alex is probable to leave.

It would be a mistake to question this synonymy. (6) is synonymous with (5), though it is ungrammatical. In fact, over the years (rightly I think), Jim has stressed the importance of the corrections that native speakers (modulo considerations of politeness) suggest to non-native speakers (or writers) toward saying correctly what they want to say. If this is what you want to say, then this is the way to say it. Ungrammatical expressions may well have perfectly clear meanings. (See also Chomsky’s old example, to the same effect: *The child seems sleeping). So clear, indeed, that the native speaker instantly finds the grammatically correct way of expressing that meaning.

Compositionality, in order to be an interesting hypothesis, has to be a locality condition on semantics.

Moreover, semantics has to be parametrized. It’s a system with a small set of restricted choices of semantic composition-combination. See the example of the English there and the Italian ci.

English and Italian have different indefiniteness conditions for their respective versions of there-insertion. There is John in the garden only has a “list” interpretation (John and Mary and Tom, or John but not Mary). The Italian C’è Gianni in giardino has the event-existential interpretation (It so happens that John is in the garden). The choice is between tracing a lexical distinction between there and ci, with different lexical selections, and admitting a difference in combinatorial powers between the two languages. Simple (so to speak) syntactic differences, with no semantic import, would amount to a failure in compositionality. The difference in meaning would be inexplicable. But the lexical distinction would be totally arbitrary. So, semantic compositionality must admit of parametric differences. Strict universality would lead to cheating. I'm almost sure I disagree with this. Hasn't anyone done some good syntactic analysis on this Italian sentence?

Follows a long and subtle analysis of conditionals as prima facie violations of compositionality (the tip of a large iceberg). Compositionality is restored via complex syntactic, logical and semantic considerations.

Other apparent exceptions to compositionality emerge in the interpretation of deontics, in many languages. These appear (much in tune with Fodor’s critique, but leading to a different conclusion) not to be compositional.

John may not leave

The meaning is that John is denied permission to leave. But the constituent structure is

[may [not leave]]

and shotgun compositionality would give the interpretation that John is given permission not to leave. Various remedies can be concocted (different interpretations of the negation, special features of deontics, higher order logic). The general lesson is:

“In formulating a restrictive semantic theory, we face a problem not in one or two

unknowns, but three. There is, first of all, the problem of saying precisely what the

meaning of a sentence or other expression actually is, or more exactly what must be

known by the native speaker who is said to know the meaning. Second, there is the

question what the inputs to semantic interpretation are: single structures at LF, or

possibly complexes consisting of LF structures and others, and so forth. And third, there

is the question what the nature of the mapping from the inputs, whatever they are, to the

interpretation, whatever it may turn out to be. In the inquiry, therefore, we must be

prepared for implications for any one of these questions deriving from answers to the

others”.

Compositionality is a “working hypothesis” that has proved to be a good hypothesis. It leads to interesting and hard choices on where to trace certain crucial dividing lines (the lexicon, syntax, parametrization versus universality, switched heads etc.)

This difficult paper deserves a closer look. Maybe next Wednesday (if there is enough consensus in this class)

Tentative conclusions:

Higginbotham is arguably offering the arguments and the data that Fodor wanted. He agrees with Fodor in considering semantic compositionality (in a restricted sense) an empirical hypothesis. He considers the arguments persuasive and the data supportive, though Fodor does not. After having maintained for years that there are no semantic parameters (only syntactic ones) and that the mapping of LF to interpretation is universal, Jim now introduces semantic parameters (headedness being the central one). Or, at any rate, parametric variations in the mapping to interpretation. It remains (to me at least) problematic how the child can learn the values of a semantic parameter. He also introduces (in other recent papers) a role for knowledge of the context. Something close to pragmatics. These components are learned, but, Jim claims, the distinction between semantics and pragmatics is clear enough to avoid all confusion.

Essential references:

Chomsky, N. (1972). Remarks on nominalization. In Chomsky, N. (Ed.), Studies on Semantics in Generative Grammar (pp. 11-61). The Hague: Mouton.

Chomsky, N. (1985). The Logical Structure of Linguistic Theory. Chicago, Ill.: The University of Chicago Press.

Chomsky, N. (1995). Language and nature. Mind 104, pp. 1-61.

Hale, K., & S. J. Keyser. (1993). On argument structure and the lexical representation of semantic relations. In Keyser, S. J. & K. Hale (Eds.), The View from Building 20. Cambridge, MA: The MIT Press.

Hale, K., & S. J. Keyser. (2002). Prolegomenon to a Theory of Argument Structure. Cambridge, MA: The MIT Press.

Higginbotham, J. (1987). The autonomy of syntax and semantics. In Garfield, J. L. (Ed.), Modularity in Knowledge Representation and Natural-Language Understanding (pp. 119-131). Cambridge, MA: Bradford Book/The MIT Press.

Higginbotham, J. T. (1985). On semantics. L.I. 16(4), pp. 547-593.

Higginbotham, J. T. (1989). Elucidations of meaning. Linguistic and Philosophy 12(3), pp. 465-517.