Neural heuristics for scaling constructional language processing

Constructionist approaches to language make use of form-meaning pairings, called constructions, capture all linguistic knowledge that is necessary for comprehending and producing natural expressions. Language processing consists then in combining the constructions a grammar such way they solve given comprehension or production problem. Finding an adequate sequence constitutes search problem combinatorial nature becomes intractable as grammars increase size. In this paper, we introduce neural methodology learning heuristics substantially optimise processes involved constructional processing. We validate case study CLEVR benchmark dataset. show our novel outperforms state-of-the-art techniques terms size space time computation, most markedly direction. The results reported on paper have potential overcome major efficiency obstacle hinders current efforts large-scale construction grammars, thereby contributing development scalable systems.