A Compound Algorithm

Compounds such as `stone lion', `vodka-juice' and `conference legs' illustrate three different puzzles in computational semantics: (i) privative constructions, (ii) headless constructions and (iii) analogy. Few existing compounding theories can solve any of these puzzles. This paper presents a theory of interpretation that is more flexible than such theories, but at the same time, computationally relatively simple.

Our treatment is based on a thorough theoretical and typological study of compounds [1]. In that study, compound constituents came with ontological information, and the meaning of compounds was compositionally derived from constituents and some schematic, bridging construction. Here we simplify matters somewhat; rather we assume type shifting operations to apply to templates subsuming the relevant constituents. For template specification we annotate individuals and events with semantic types.

Our approach leads us to treat privative constructions as type-lifted intersective constructions, treat compounds as a priori headless constructions (cf. cross-linguistic evidence) and allow lexical constituents to form analogy bases.

The final part of the paper illustrates how our theory applies to machine translation, and it briefly sketches how its domain may be extended to include other kinds of noun phrase modifiers.

References

[1] Soegaard, A. (in press). Compounding Theories and Linguistic Diversity. In: Frajzyngier (ed.), Language Theories and Linguistic Diversity, John Benjamins, Amsterdam.