Séminaire de Recherche en Linguistique

 

In this talk, I will present an interdisciplinary study between theoretical and computational

linguistics on the interpretation of deverbal compounds. The deverbal compounds in our

study are noun-noun compounds whose head noun is derived from a verb, for example

border protection, police protection, or fire protection. The non-heads of these compounds

(i.e., border, police, fire) may be understood as objects or other dependents of the original

verb (cf. to protect a border; the police protect the people; to protect from fire). The focus of

this work is on determining to what extent the morphosyntactic and lexico-semantic

properties of the deverbal noun head (i.e., protection) can help in predicting the

interpretation of the compound, as mediated by the syntactic-semantic relationship that the

head establishes with the non-head (here, object vs. non-object). 

 

The starting point is Grimshaw‘s [1] observation that deverbal nouns are ambiguous

between compositionally interpreted Argument Structure Nominals (ASNs), which inherit

verbal event structure and realize arguments (e.g., the protection of the border by the

police), and more lexicalized Result Nominals, which preserve no event structure or

arguments (e.g., The mask is a protection for his nose). Under this view, we expect the non-

heads of deverbal compounds whose heads share many properties with ASNs (in other

words, heads that have a high level of ASN-hood) to be interpreted more often as direct

objects than deverbal compounds that do not have these properties. This observation

contrasts with Borer’s [2] view that deverbal compounds have no arguments or event

structure, which predicts no correlation between the ASN-hood of the head and the

interpretation of the non-head as a direct object.

 

We conducted experiments with simple machine learning models to test the predictive power

of features indicative of ASN-hood as presented in linguistic literature, and combined

frequency counts from a large corpus with evidence from human judgements. We found

support for Grimshaw’s hypothesis, however, the study raises interesting discussion points

on how relevant the different properties of the head are for the interpretation of the deverbal

compound, and on the reliability of corpus-based features as proxies for the level of  ASN-

hood. 

 

[1] Grimshaw, Jane (1990) Argument Structure. MIT Press, Cambridge, MA

[2] Borer, Hagit (2013) Taking Form. Oxford University Press, Oxford