Induction of Recursive Transfer Rules
Transfer rules are used in bi-lingual translation systems for transferring a logical representation of a source language sentence into a logical representation of the corresponding target language sentence. This work studies induction of transfer rules from examples of corresponding pairs of source-target quasi logical formulae (QLFs). The main features of this problem are: i) more than one rule may need to be produced from a single example, ii) only positive examples are provided and iii) the produced hypothesis should be recursive. In an earlier study of this problem, a system was proposed in which hand-coded heuristics were employed for identifying non-recursive correspondences. In this work we study the case when non-recursive transfer rules have been given to the system instead of heuristics. Results from a preliminary experiment with English-French QLFs are presented, demonstrating that this information is sufficient for the generation of generally applicable rules that can be used for transfer between previously unseen source and target QLFs. However, the experiment also shows that the system suffers from producing overly specific rules, even when the problem of disallowing the derivation of other target QLFs than the correct one is not considered. Potential approaches to this problem are discussed.
Unable to display preview. Download preview PDF.
- 1.Alshawi H. (ed.) (1992). The core language engine, Cambridge,MA: MIT PressGoogle Scholar
- 2.Agnäs, M-S., Alshawi, H., Bretan, I., Carter, D., Ceder, K., Collins, M., Crouch, R., Digalakis, V., Ekholm, B., Gambäck, B., Kaja, J., Karlgren, J., Lyberg, B., Price, P., Pulman, S., Rayner, M., Samuelsson, C. & Svensson, T., (1994). Spokenlanguage translator: first-year report, SICS research report, ISRN SICS-R-94/03-SE, StockholmGoogle Scholar
- 3.Boström, H., (1998). Predicate Invention and Learning from Positive Examples Only. In Proceedings of the Tenth European Conference on Machine Learning, Berlin Heidelberg: Springer-Verlag (pp. 226–237)Google Scholar
- 4.Boström, H., & Zemke, S., (1996). Learning transfer rules by inductive logic programming (preliminary report), Dept. of Computer and Systems Sciences, Stockholm University and Royal Institute of TechnologyGoogle Scholar
- 5.Clocksin, W.F., & Mellish, C.S., (1981). Programming in prolog, Springer VerlagGoogle Scholar
- 6.Cussens, J., (1999). Loglinear Models for First-Order Probabilistic Reasoning. In Proceedings of Uncertainty in Artificial Intelligence, San Francisco: Morgan Kaufmann (pp. 126–133)Google Scholar
- 7.Lloyd, J. W., (1987). Foundations of logic programming, (2nd edition), Berlin Heidelberg:Springer-VerlagGoogle Scholar
- 8.Milward, D., & Pulman, S., (1997). Transfer learning using QLFs, Technical report, SRI International, Cambridge, G.B.Google Scholar
- 9.Mooney, R. J., & Cali., M. E., (1996). Learning the past tense of English verbs using inductive logic programming. In S. Wermter, E. Rilo. & G. Scheler (eds.), Symbolic, Connectionist and Statistical Approaches to Learning for Natural Language Processing, Berlin Heidelberg: Springer-Verlag (pp. 370–384)Google Scholar
- 11.Muggleton, S., (1995). Stochastic logic programs. In De Raedt L. (ed.), Advances Inductive Logic Programming, Amsterdam: IOS Press (pp. 254–264)Google Scholar
- 12.Quinlan, J. R., (1990). Learning Logical Definitions from Relations. In Machine Learning 5 239–266Google Scholar