How the meaning of a compound word like "doghouse" arises from "dog" and "house" is a long-standing puzzle in lexical semantics. This paper tackles it by comparing computational models that predict compound meanings from their constituent parts.
The authors argue that comparing models provides a way to evaluate alternative accounts of compound interpretation. Their focus is on analogy: the idea that the meaning of a new compound is derived through analogical mapping from known compounds.
Because the abstract is brief, the specific models, datasets, and results are not detailed here. The work points to analogy as a promising explanatory principle, but the strength of the evidence will depend on the full paper.