Semantics: Notes 3

Emmon Bach, SOAS, UMass(Amherst)
Oxford: 29 January, 2008
contact: ebach@linguist.umass.edu
Copyright Emmon Bach 2008. All rights reserved.
Link to Notes:                                            
"http://www.people.umass.edu/ebach/courses/ox08-pl.htm"

(29 January) Fragment I; Linguistic ontology: what do we talk as if there is?

Preclass: discussion of Exercises from Exx-2.

Fragment I: A simple language L1

We will look first at a complete fragment for a very small subset of written English, following the formalism and style of Montague's PTQ. We will reflect a bit on the properties of this grammar. Then we will recast the grammar in a categorial framework.

Thought questions:
  1. What problems will we encounter when we add third person pronouns to our language and grammar?
  2. What problems will we encounter when we add plurals to our language and grammar?
  3. What problems will we encounter when we add negation to our language and grammar?
  4. What problems will we encounter when we add first and second person pronouns to our language and grammar.
Comments on the grammar so far: like the semantic interpretation, the grammar just given follows Montague's lead in being firmly rooted in set theory. That is, it defines sets of expressions. You need to be clear about this if you compare this sort of grammar with the phrase-structure grammars and their descendents familiar from the generative tradition. What the latter generate are representations of structures: phrase-structure diagrams, phrase-markers and so on. There is a natural correspondence between the derivation of a member of a complex category here and the phrase-structure trees of TG and other theories, but it is deceptive. The closest correspondent in early TG was the notion of a T-marker (as opposed to a P[hrase]-marker). A T-marker was a collection of P-markers and a record of their transformational history. [Illustrate on board.]

(using j and m to stand for John and Mary)

            John loves Mary,S  :   [love'(m)](j)
                /\
               /  \
              /    \
             /      \
       John,N : j  loves Mary,IV : love'(m)
                          / \
                         /   \
                        /     \
                       /       \
                      /         \
             loves, TV : love'  Mary, N : m

This object is an analysis tree, of which more below, annotated with the syntactic category of the string so far, and a representation of the denotation of the expression. We will continue to use such representations. We will also take up the question of the theoretical status of the representation of the semantic value of the exressions.

Montague's general theory was set forth in all its gory formal details in his "Universal Grammar" (1970b). There a clear distinction was made between rules and operations. Each rule specifies a sequence of k categories (the arguments), another category (the result), and a k-place operation.

What is an operation? A function from some set to that same set. What is the set in this case: the set of sequences of strings of symbols in the language of the grammar. So in general, the operations of the grammar are functions from sequences of strings of symbols to (one member) sequences of strings of symbols. This set -- call it Σ -- is a superset of the set of well formed written strings of symbols of English. So the grammar is a way of picking out the subsets of Σ that are members of various syntactic categories of the language.

Syntax and Semantics
A syntactic function or functor is an expression that takes an expression of a certain class and by combining with it by some operation yields an expression of another class.

So a functor specifies an argument category, a result category, and an operation. More generally, n argument categories, a result category and an n-place operation. We restrict our attention here to 2-place functors. In the familiar categorial notation we use categories like this:

(More on this below. )The rule-to-rule assumption about the relation between syntax and semantics requires that every syntactic rule specifies the interpretation of the result of the rule as a function of the interpretation of the inputs to the rule. So the syntax and semantics operate in parallel. This is our implementation of the Fregean principle of compositionality:

The interpretation of the result of a syntactic rule is a function of the interpretation of the expressions combined and the manner of their combination.
Categorial Grammar
A particularly elegant and appealing approach to the specification and interpretation of a language that mirrors the type theory of the interpretation is that of a categorial grammar.

Definition of CAT:

  1. N, S are members of CAT
  2. if a is a member of CAT and b is a member of CAT, then so are a/b and a\b.
  3. Nothing else is a member of CAT
This is a highly abbreviated and oversimplified kind of syntax. We interpret the notation like this: expressions in a category a/b (a\b) take arguments of category b to their right (left) to form expressions of category a.

Warning! There are several notations for categorial grammars kicking around. In this one (following Steedman 2000 and others) the argument category is always on the right. In another popular notation the argument is always "under" the slash, so where we have here a/b the other notation would have b\a.
Now we can straightforwardly specify the relation between the syntax and the semantics. In an interpretation that ignores intensionality we would have the function m relating categories and types like this: That is, functors are uniformly interpreted as functions from denotations of the type corresponding to the category of the argument to denotations of the type corresponding to the category of the result.

When intensions are brought in then these simple correspondences are extended to include the parameter of a possible world.

What we have so far is essentially the same setup that is used by Montague in his most famous paper on natural language (PTQ = Montague 1973), except that that paper uses a set of times as well as a set of worlds.

A Categorial Grammar for L1:

Now rather than stating individual rules we state once and for all rules for functional application to the right (FA>) and to the left FA<. In line with the rule-to-rule constraint on interpretations, we simply invoke the mappings defined above and write the interpretation rule as part of the schema, separated the syntax from the semantics with a colon:
If α ∊ Pa/b and β ∊ Pb, then αβ ∊ Pa and D(αβ) = D(α)(D(β)). Abbreviated:
  • FA>: a/b b ==> a : D(a/b)(D(b))
  • If β ∊ Pb and α ∊ Pa\b, then βα ∊ Pa and D(βα) = D(α)(D(β)). Abbreviated:
  • FA<: b a\b ==> a : D(a\b)(D(b))
    1. Thought question: what additional formal machinery can you think of that we would need to give an explicit account of the grammar of a natural language like English?

      History: Categorial grammars were invented by the Polish logician and philosopher Kasimierz Ajdukiewicz (1935), although the basic idea has roots in work by Łukasiewicz, Frege, and ultimately Aristotle. Ajdukiewicz's formulation used bidirectional functors, built from the basic categories n and s (think: names and sentences). Functors are written as fractions (for convenience here we may also use |), so we have:
      FA:         a              a
                 ___   b  =  b  ___  ==>  a
      
                  b              b
      
      Or:
             a|b b = b a|b ==> a
      
      (Again: argument on the right!)
      
      Later, several linguists and logicians took up Ajdukiewicz's ideas, most notably Yehoshua Bar-Hillel and Joachim Lambek, who used the slashes / and \ to indicate right and left concatenation respectively. At the conference at Stanford that went into Davidson and Harman (1972), Peter Geach made creative use of Ajdukiewicz's ideas. A sign of the times: Geach's paper was called "A programme for syntax" and at the same conference Jim McCawley gave a paper called "A program for logic." The two papers went right past each other.

      Joachim Lambek published several papers on categorial systems around 1960 (Lambek 1958, 1961), as did Haskell Curry (1961). Sir John Lyons (1966) considered categorial grammars as a base for transformational grammar, as did David Lewis (1972). Lyons' work was an important step toward X-Bar theory (Jackendoff 1977) and in turn drew on questions and ideas of Zellig Harris (1951)). Lyons asked whether the occurrence of "N" on both sides of rules like "NP --> Det N" was more than a pun.

      Categorial grammars of the directional type were early shown to be weakly equivalent to context-free (CF) grammars. At the time many people bought the supposed demonstration by N. Chomsky of the inadequacy of context-free grammars for natural language and hence directional categorial grammars, so interest died out (with the notable exceptions mentioned in the last paragraph). Another honourable exception: Gilbert Harman 196? -- FILL IN! -- his paper may well be considered a precursor to Gazdar etc. to be mentioned below for realizing the importance and possibilities of feature systems. Chomsky's dismissive and polemical response may be found in Chomsky (FILL IN).

      Natural language ontology: what do we talk as if there is?

      To be is to be the value of a variable.
      Willard Van Orman Quine "On what there is"

      Philosophers, physicists, and other people worry about what the ultimate nature of reality is. They ask the question:

      What is there?

      Linguists have the luxury of asking instead the question:

      What do people talk as if there is?

      We might call this sort of investigation natural language metaphysics or natural language ontology.

      In this session, we will just raise some first questions about this enterprise. Later in the course, we will come back to such question when we deal with the option of sorting the domain, or dealing in some other way with the different kinds of things that natural languages seem to distinguish in one way or other.

      The Whorfian question: Do different languages require, induce, or reflect different views of reality?

      Ordinary people, not just philosophers or anthropologists seem to think so. Can linguistic semantics help us to think about such questions?

      What have we got in our model structures so far?

      The set of individuals A: no structure, no differentiation of sorts, etc. Later we will look at suggestions for structuring the domain by adding Sorts for e.g. Kinds and (ordinary) Objects and Stages, Plurals, Properties as a basic Sort, Events of various kinds and so on. Formally, this might mean just dividing A into subsets. One effect of this is to make it possible to distinguish out and out "unsemantic" sentences (a) from "sortally incorrect" sentences (b):

      1. John is loves.
      2. This sentence would also be ungrammatical, i.e not derivable from the syntax.
      3. Max is widespread.
      4. Compare: Ants are widespread.

      Note on multiple explanations.

      Exercises 2: on functions and lambdas

      Discussion of Exercises 2

      References:
      Categorial grammars:
      Introduction to Oehrle et al. 1988
      Oehrle et al. 1988
      Bach 1988, 2005
      Steedman 2000.

      Mathematical Structures for Semantics: Landman 1991

      Compositionality: Partee 1984
      A good general discussion of compositionality online in the Stanford Encyclopedia of Philosophy

      Natural Language Metaphysics: Quine 1948, Bach 1981, 1986