Parsing Clinical Finnish: Experiments with Rule-Based and Statistical Dependency Parsers

In this paper, we present a new syntactically annotated corpus consisting of daily notes from an intensive care unit in a Finnish hospital. Using the corpus, we perform experiments with both rule-based and statistical parsers. We apply an existing rule-based parser specifically developed for this clinical language and create a set of conversion rules for transforming the constituency scheme of this parser into the dependency scheme of the corpus. The statistical parser is induced from the corpus using the MaltParser system. We find that even with a modestly-sized corpus, the statistical parser achieves results comparable to those previously reported on a number of languages using considerably larger corpora. The accurate constituency-to-dependency conversion improves the applicability of the rule-based parser by inferring grammatical roles, thus deepening its analyses.