INCREMENTAL PREDICTION OF SENTENCE-FINAL VERBS WITH ATTENTIVE RECURRENT NEURAL NETWORKS

Abstract

Sentence-final verb prediction has garnered attention both in computational lin- guistics and psycholinguistics. It is indispensable for understanding human processing of verb-final languages. More recently, it has been used for computational approaches to simultaneous interpretation, i.e. translation in real-time, from verb-final to verb-medial languages. While previous approaches use classical statistical methods including pattern- matching rules, n-gram language models, or a logistic regression with linguistic features, we introduce an attention-based neural model, Attentive Neural Verb Inference for Incre- mental Language (ANVIIL), to incrementally predict final verbs on incomplete sentences. Our approach better predicts the final verbs in Japanese and German and provides more interpretable explanations of why those verbs are selected.