ROLAQUAD: Robust language understanding for question-answering dialogues
We propose basic research in the context of a multiple-turn question-answering dialogue application focusing on the following problems:
- The integration of multiple levels of information (question understanding, move diagnosis, topic detection).
- Robustness of processing in the presence of many interacting modules (e.g. the shallow understand ing and speech recognition modules) presenting partial an d noisy information.
The method we propose for solving these problems is a machine learning framework in which the different modules of the system are considered as instances of feature-rich classification tasks that are trained on example data. Recent machine learning techniques such as Conditional Random Fields and Maximum Entropy Ma rkov Models will be applied and further developed. Although the focus will be on the former issues, the project will also address move generation and answer generation from a knowledge base by question-answer alignment. The proposed approach is language-independent, and will be tested on English and Dutch data.
Projectleider: Prof. dr. A.P.J. van den Bosch (UvT)
Project website: http://ilk.uvt.nl/rolaquad/