Speaker: Pauline Jacobson (Brown University)
Title: A Categorial Grammar view of syntactic categories: Some happy coordination surprises
Time: Thursday, May 4th, 12:30pm – 2pm
Location: 32-D461
Abstract: While there are various versions of Categorial Grammar (CG); this talk is rooted in one maintaining the following: (1) syntactic categories are just encodings of the syntactic distribution of any item and its semantic type, (2) the architecture of the grammar is Direct Compositional: the syntax ‘builds’ (defines as well formed) expressions while the semantics simultaneously assigns a meaning (a model-theoretic object) to each expression proven well formed in the syntax. The rules often combine two expressions to give a larger one, though there are also unary rules taking a single expression (a triple of <[sound], Cat, meaning>) as input and giving a triple as output.
I will explore a view that takes seriously the notion that not only are the meanings of expressions often functions from one set to another, but that syntactic categories also correspond literally to functions: these apply to one string, then another to give a third. (E.g., the function corresponding to the category of fix is such that when applied to the string fix it gives a function from the string the sink to fix the sink.) See Jacobson, Compositional Semantics: An Introduction to the Syntax/Semantics Interface OUP 2014 Chapter 6 for useful background. While this sounds rather legalistic, we will see its benefit when combined to notions like function composition and type lifting. Thus, e.g., Steedman (1984) and Dowty (1987) have shown that adding function composition (to be “Curry’ed here) and type/category lifting into the inventory of syntactic operations gives elegant accounts of, e.g., Right Node Raising examples (Lee loves and Sandy hates model theoretic semantics) and other kinds of so-called ‘nonconstituent coordination’ as in Cap’n Jack featured lobster yesterday and scallops today.
One further piece is the analysis of and. Here and in many other theories it is a normal lexical item taking an expression of category X first to its right and then another X to its left to give an X. (This is not news – many theories agrees on this.) All other things being equal, one would not expect there to be a Coordinate Structure Constraint and indeed many have argued that the CSC effects are pragmatic. While I have nothing new to offer on just what gives rise to these effects, I will agree with Lakoff (1984) that violations of the CSC in ‘silent coordination’ chains supports the hypothesis that CSC effects are not syntactic Lakoff’s key examples are things like How much beer did John buy __ , pack up __, bicycle home [no gap here!] and then proceed to drink __? (Note: one can also construct examples without the ‘and then’ interpretation.) I will give a rough analysis of the ‘silent’ and and show interesting results of the Lakoff chains with respect to Heavy NP Shift/RNR cases. For example it follows that in a Lakoff chain there can be a mix and match of where there are gaps and where not provided that there is a gap on the rightmost conjunct in “right movement” cases: John bought __, packed up __, bicycled home, and drank __ a full 2 sixpacks of beer vs. *John bought, packed up __, drank __, and fell asleep a full 2 sixpacks of beer. (In left “extraction” – as in the actual Lakoff example – this pattern is different for reasons that I will at best briefly discuss.) The results crucially depend on the notion of syntactic categories as corresponding to actual functions from strings to (strings to strings).
The punchline comes asking: Is there is anything analogous on the left? And indeed – and spectacularly – there is. Thus Maxwell and Manning (1996) noted the surprising existence of cases like the following: David teaches Minimalism in the morning, HPSG in the afternoon, and plays the violin in the evening. The existence of these follows automatically from the machinery assumed here. Moreover it follows that the fuller expression must be rightmost, as in *David teaches Minimalism in the morning, plays the violin in the afternoon, and HPSG in the evening. In fact, the full set of ‘mix and match’ is more constrained than for the right cases above, due to the asymmetry of and. The empirical facts happily match up with the predictions.