Institut de Lingüística Aplicada

Third Mediterranean Meeting on Morphology (MMM3)


Rochelle Lieber (University of New Hampshire)
Compound interpretation: lexical semantics, not syntax

In this paper I will reconsider the derivation and interpretation of both synthetic compounds (truck driver, cat loving, hand made, cost containment, etc.) and root compounds (file cabinet, towel rack, producer-director, red cap) in English. Specifically, I will argue against past analyses that explain compound interpretation either as a matter of the internal syntactic structure of compounds (Lieber 1983) or of the operation of syntactic movement rules (Roeper 1988, Lieber 1992). I will elaborate on a theory of lexical semantic representation sketched in Lieber & Baayen 1999, and show how it offers natural explanations for the following often-made observations about compound interpretation:

The proposed analysis takes the task of compound interpretation out of the syntax and thus strengthens the autonomy of the morphological component.

While the internal structure of root compounds has never been controversial (1), two different internal structures have been proposed for synthetic compounds (2):

  1. [[ ]X [ ]Y]Y
  2. a. [[ ]X [[ ]Y afx]Z ]Z
    b. [[[ ]X [ ]Y]Y afx]Z

The analysis in (2a) is the most widely accepted (see, for example, Selkirk 1982, Booij 1988, 1992), although structure (2b) has been argued for as well (Lieber 1983). Significantly, the arguments for the analysis in (2b) often hinge on questions of semantic interpretation. Further, issues of interpretation have sometimes driven researchers to use syntactic devices such as transformations and movement rules in the analysis of synthetic compounds (e.g., Roeper and Siegel 1978, Roeper 1988, Lieber 1992). I will argue that root compounds have the structure in (1), and synthetic compounds the structure in (2a), and that no syntactic rules or principles are necessary to account for compound interpretation. Specifically, I will argue two points in this paper:

The internal structure of compounds should not be driven by considerations of interpretation. Instead, interpretation must be accounted for within a well-articulated theory of lexical semantics. Nevertheless, the mechanisms available within that theory of lexical semantics should make it possible to explain the interpretation of both root compounds (and among them not only ordinary endocentric ones, but also the so-called dvandva and bahuvrihi compounds) and synthetic compounds in a uniform way.

Elaborating on the framework of lexical semantic representation outlined in Lieber & Baayen 1999, I will propose that compound formation involves the juxtaposition of two lexical semantic representations, and the linking (co-indexing) of an argument of the first representation with one of the second. Linked arguments share reference and interpretation; that is, if the argument of the second stem receives an interpretation such as agent or patient by virtue of the semantic representation of that stem, that interpretation is passed on to the argument which shares its index. Schematically, the semantic representations of root compounds look like (3), and those of synthetic compounds like (4), where the feature [+substance] indicates, roughly speaking, a concrete noun, [+substance, -dynamic] an agent or instrument noun, and [+dynamic] a simple activity verb:

  1. e.g., file cabinet
    [+substance ([ ]i)] [+substance ([ ]i)]
    (file) (cabinet)
  2. e.g., truck driver
    [+substance ([ ]j)] [+substance, -dynamic ([ ]i) [+dynamic ([ ]i, [ ]j)]]
    (truck) (-er) (drive)

The first stem of compounds must be nonreferential in interpretation because its reference is always linked to another argument in the compound; in other words, its reference is always dependent on the reference of another argument in the compound structure. In synthetic compounds, the nominalizing affix binds one of the arguments in the lexical semantic representation of the base verb (for -er it is the first verbal argument). The first stem in the compound must then be co-indexed with a free argument, which in the case of most synthetic compounds is the second verbal argument. Since this argument in most cases receives the patient interpretation, the first stem receives that interpretation as well. I will show that the simple mechanisms of juxtaposition and co-indexing in effect explain generalizations about compound interpretation that have been variously codified as the "IS A Principle" (Allen 1978), the First Sister Principle (Roeper and Siegel 1978), the First Order Projection Condition (Selkirk 1982), and the Argument Linking Principle (Lieber 1983).