Coercion and Copredication as Flexible Frame Composition is the title of a DFG-funded research project lead by Laura Kallmeyer and Rainer Osswald that is being conducted at Heinrich Heine University Düsseldorf from 02/2021 to 02/2024.
Theories of semantic composition have to cope with the flexible semantic behavior of words that often shows up when they co-occur in syntagmatic relationships. The project addresses two such phenomena. The first one is concerned with the varying meaning contributions of inherently polysemous words depending on the predications in which they occur. Inherent systematic polysemy means that a word or phrase has two or more clearly distinguishable but related interpretations and that there are other words or phrases which show the same pattern of semantic variation. Well-known examples are nouns like “book” and “letter”, which can denote physical or informational objects, and “lunch” and “dinner”, which can refer to food or events. It is characteristic of these nouns that they can occur in copredication constructions, as in “memorize and burn the letter”, where two or more predicates that aim at different meaning facets are jointly applied to the same argument. The second phenomenon is the systematic coercion of semantic enrichment that occurs if a mismatch between the semantic type of an argument expression and the requirements of the predicate is not resolved by facet selection, but by an extended mode of composition in which appropriate pieces of information are added to eliminate the mismatch. This is the case, for instance, when a phase verb such as “finish” or a psych verb such as “enjoy” takes a non-eventive argument such as “letter” or “coffee”. The goal of the project is to develop a frame-semantic account of these phenomena which combines rich semantic representations in the lexicon with a formal model of flexible compositionality at the interface between syntax and semantics. The underlying hypothesis is (i) that a frame-based semantic decomposition is particularly well suited for modeling the various meaning facets of polysemous words and the relations between them, and (ii) that an architecture of the syntax-semantics interface that links semantic frames to the elementary syntactic trees of Lexicalized Tree Adjoining Grammars gives us the right amount of flexibility for modeling copredication and coercion constructions. Moreover, the formal analyses developed in the project will be based on a broad empirical basis which includes copredication and coercion data semi-automatically extracted from corpora.