|zpravodaj ČSKI - leden 2000 [ pdf ]|
datum: 10.1.2000 v 11:00
název: Learning parsers from Annotated Corpora
přednášející: Dimitar Kazakov (GL laboratoř, na stáži v Department of Computer Science, University of York)
místo konání: CVUT, 1.patro, místnost č.112
souhrn: The theme of this seminar will be LAPIS or ILP learning of specialised LR parsers: Extracting backbone CFG from the corpus. Generation of an LR parser. Rewriting the parser in Prolog. Construction of learning examples. Learning of specialised parsers. Results.
Speech recognition and LAPIS: IBM VoiceType. Spoken Language Understanding Prototype. Lapis for SRCL grammar induction. Ongoing & Future Work. Conclusions.
Bonus track (according to the time left)
Positive-only learning and analogical prediction in Prolog(4.4):
- Pos-only learning is based on a trade-off between the generality of the theory learned and the length of its binary representation. The relationship between pos-neg and pos-only learning of multi-class classification will be discussed from a practical point of view.
- Analogical prediction is a kind of lazy learning where a separate 1st-order theory is created for the classification of each test example. The underlying principle will be explained.