Bottom-up parsing approach to modelling local coherence effects

Abstract

In the field of psycholinguistics, sentence interpretation has huge discussions due to the level of ambiguity. Human sentence processing occurs incremen- tally. The central question of grammatical constraints lies with the levels of ambiguity. Several theories and parsing models have been tested to attack distracting effects of wrongly interpreted sentences. Previous Noun Phrase- Verb Phrase sequences have been tested on human reading times and found local syntactic coherence effects. This paper asks if a bottom-up transition- based parsing model can predict activation numbers that compare to human results in order to further support the theory of local coherence effects. We ran 20 sentences in 4 different conditions through a bottom-up parser and used mean activations to showcase the effects of local ambiguity. Our data shows the results are consistent with the hypothesis and show promising re- sults for further research in modelling local coherence effects.