On deriving semantic representations from dependencies: A Practical approach for evaluating meaning in learner corpora

One of the reasons for the popularity of dependency approaches in recent computational linguistics is their ability to efficiently derive the core functorargument structure of a sentence as an interface to semantic interpretation. Exploring this feature of dependency structures further, in this paper we show how basic dependency representations can be mapped to semantic representation as used in Lexical Resource Semantics (Richter and Sailer 2003), an underspecified semantic formalism originally developed as a semantic formalism for the HPSG framework (Pollard and Sag 1994) and its elaborate syntactic representations. We describe a two stage process, which in the first stage establishes a syntaxsemantics interface representation abstracting away from some differences in surface dependencies. It ensures the local reconstruction of arguments for middle and long-distance dependencies, before building the actual LRS semantics in the second stage. We evaluate the approach on the CREG-109 corpus, a small dependency-annotated corpus with answers to reading comprehension questions written by American learners of German.