This paper presents a new memory-bounded left-corner parsing model for unsupervised raw-text syntax induction, using unsupervised hierarchical hidden Markov models (UHHMM). We deploy this algorithm to shed light on the extent to which human language learners can discover hierarchical syntax through distributional statistics alone, by modeling two widely-accepted features of human language acquisition and sentence processing that have not been simultaneously modeled by any existing grammar induction algorithm: (1) a left-corner parsing strategy and (2) limited working memory capacity. To model realistic input to human language learners, we evaluate our system on a corpus of child-directed speech rather than typical newswire corpora. Results beat or closely match those of three competing systems.

Convolution tree kernels are an efficient and effective method for comparing syntac- tic structures in NLP methods. However, current kernel methods such as subset tree kernel and partial tree kernel understate the similarity of very similar tree structures. Although soft-matching approaches can im- prove the similarity scores, they are corpus- dependent and match relaxations may be task-specific. We propose an alternative ap- proach called descending path kernel which gives intuitive similarity scores on compa- rable structures. This method is evaluated on two temporal relation extraction tasks and demonstrates its advantage over rich syntactic representations.