The Interplay of Syntax and Morphology in Building Parsing Models for Modern Hebrew

As of yet, there is no statistical parser for Modern Hebrew (MH). Current practice in building parsing models is not immediately applicable to languages that exhibit strong interaction between syntax and morphology, e.g. Modern Hebrew, Arabic and other Semitic languages. We suggest that incorporating morphological and morphosyntactic information into the parsing model is essential for parsing Semitic languages. Using a morphological analyzer, a part-of-speech tagger, and a PCFG-based general purpose parser, we segment and parse unseen MH sentences using a small annotated corpus. The Parseval scores obtained are not comparable to those of, e.g., state-of-the-art models for English, due to remaining syntactic ambiguity and limited morphological treatment. We conjecture that adequate morphological and syntactic processing of MH should be done in a unified framework in which morphology and syntax can freely interact and share information in both directions.