Lexical-semantic factors in event interpretation
Principal Investigator: Sebastian Padó, Sabine Schulte im Walde
Researchers: Alessandra Zarcone, Jason Utt
Associated Ph.D. Student: Gabriella Lapesa
Project D6 was concerned with a particular phenomenon of incremental specification in context, namely the interpretation of verb-object pairs. While some verb-object pairs are interpreted compositionally ("simple composition", as in begin a holiday), the understanding of others involves covert events that are not realised on the surface ("enriched composition": for example, begin a song, in neutral context, implies begin singing a song).
The interpretation of such pairs involves two specification steps:
(1) the decision between simple and enriched composition;
(2) in enriched composition, the specification of the covert event.
The project used psycholinguistic and computational techniques to elucidate the as yet neglected role of lexical-semantic context factors. Concretely, the literature about step (1) assumes that the ontological type of the object (event/object) is the main determinant. Since this is arguably an oversimplification, it collected and modeled experimental data on:
(a) the behaviour of ambiguous nominalisations and nouns;
(b) the respective influence of ontological type and plausibility;
(c) the role played by aspectual features.
In its turn, step (2) was treated by previous work as merely the choice of a single most plausible event. In contrast, the project elicited a range of covert events for each instance of enriched composition (for begin a song, this might be sing, chant, and croon), and analysed what semantic relations hold within this set. This analysis was used to model the interpretations in terms of concepts, to be approximated by verb clusters. The interpretation process will be analysed in parallel for English and Italian.
A practical result of our studies was a framework for distributional corpus-driven, broad-coverage computational models that predicts both the simple/enriched composition distinction for sentences potentially involving covert events, and, in the case of enriched composition, the range of interpretations appropriate for the present context.
Finally, it extended the scope of our core results in three ways.
(1) We phrased the recovery of covert event interpretations as an entailment task, to make our phenomenon accessible to a larger audience in computational semantics;
(2) We studied a related phenomenon, namely Italian da-nominals, which also imply covert events, e.g.
scatola da scarpe / shoe box -> box to store shoes;
(3) We developed a distributional-driven characterisation of constructions whose interpretation involves covert events.