An Application of Lexical Semantics Annotation to Question-Answering in e-Farming

In this poster we present an approach to responding to complex questions in the agriculture domain, from specifications given by experts. We present in particular a semantic annotation procedure that would allow us to define accurate and domain dedicated forms of lexical semantics inference, in order to be able to match non factoid questions (i.e. questions whose response is a significant text portion) with documents on a large scale. This project is designed to help farmers to get advices via question answering on SMS in order to improve rice farming. 1 Challenges and Goals Question answering (Moldovan 2000, Maybury 2004) operates on top of search engines of classical textual database querying tools, by providing a layer that has natural language understanding and generation as well some reasoning capabilities in order to provide users with responses which are much more accurate and cooperative than what search engines provide in general. This is particularly crucial when responses are not straightforward, e.g. when they require some form of elaboration (synthesis of data, consistency checking, etc.), reasoning or when the response is not a simple item, but a well-formed fragment of text, e.g. a chain of events leading to a consequence, a procedure, etc. The project we present here emerged from a need from the Thai Ministry of Agriculture. The main goal is to develop tools for e-Farming, in particular rice farming, so that farmers can easily get information on farming rice and rice diseases, for example via SMS. The Thai Ministry of Agriculture has