HPSG Supertagging: A Sequence Labeling View

Yao-zhong Zhang, Takuya Matsuzaki and Jun'ichi Tsujii

11th International Conference on Parsing Technology (IWPT 2009)
Paris, France, 7th-9th October, 2009

---

Summary

Supertagging is a widely used speed-up technique for deep parsing. In another aspect, supertagging comes to be exploited in other NLP than parsing for utilizing the rich syntactic information given by the supertags. However, the performance of supertagger is still a bottleneck in such applications. To improve the accuracy of supertagging, we investigated the relationship between supertagging and parsing; We started from a sequence labeling view of HPSG supertagging, examining how well a supertagger can do when separated from parsing. Comparison of two types of supertagging model, point-wise model and sequential model, showed that the former model works competitively well despite its simplicity, which indicates the true dependency among supertag assignments is far more complex than the crude first-order approximation made in the sequence model. We then analyzed the limitation of separated supertagging by using a CFG-filter. The results showed that big gains could be acquired by resorting to a light-weight parser.

---

START Conference Manager (V2.56.8 - Rev. 780)