Walter Daelemans,比利时安特卫普大学教授。Antal van den Bosch,荷兰蒂尔堡大学教授。
目录
· · · · · ·
导读
Preface
1 Memory—Based Learning in Natural Language Processing
1.1 Natural language processing as classification
1.2 A linguistic example
1.3 Roadmap and software
· · · · · ·
(更多)
导读
Preface
1 Memory—Based Learning in Natural Language Processing
1.1 Natural language processing as classification
1.2 A linguistic example
1.3 Roadmap and software
1.4 Fiirther reading
2 Inspirations from linguistics and artificial intelligence
2.1 Inspirations from linguistics
2.2 Inspirations from artificial intelligence
2.3 Memory—based language processing literature
2.4 Conclusion
3 Memory and Similarity
3.1 German plural formation
3.2 Similarity metric
3.2.1 Information—theoretic feature weighting
3.2.2 Alternative feature weighting methods
3.2.3 Getting started with TiMBL
3.2.4 Feature weighting in TiMBL
3.2.5 Modified value difference metric
3.2.6 Value clustering in TiMBL
3.2.7 Distance—weighted class voting
3.2.8 Distance—weighted class voting in TiMBL
3.3 Analyzing the output of MBLP
3.3.1 Displaying nearest neighbors in TiMBL
3.4 Implementation issues
3.4.1 TiMBL trees
3.5 Methodology
3.5.1 Experimental methodology in TiMBL
3.5.2 Additional performance measures in TiMBL
3.6 Conclusion
4 Application to morpho—phonology
4.1 Phonemization
4.1.1 Memory—based word phonemization
4.1.2 TreeTalk
4.1.3 IGTree in TiMBL
4.1.4 Experiments： applying IGTree to word phonemization
4.1.5 TRIBL： trading memory for speed
4.1.6 TRIBL in TiMBL examples Editing
4.2 Morphological analysis
4.2.1 Dutch morphology
4.2.2 Feature and class encoding
4.2.3 Experiments： MBMA on Dutch wordforms
4.3 Conclusion
5 Application to shallow parsing
5.1 Part—of—speech tagging
5.1.1 Memory—based tagger architecture
5.1.2 Results
5.2 Constituent chunking
5.2.1 Results
5.2.2 Using Mbt and Mbtg for chunking
5.3 Relation finding
5.3.1 Relation finder architecture
5.3.2 Results
5.4 Conclusion
6 Abstraction and generalization
6.1 Lazy versus eager learning
6.1.1 Benchmark language learning tasks
6.1.2 Forgetting by rule induction is harmful in language learning
6.2 Editing
6.3 Why forgetting examples can be harmful
6.4 Generalizing examples
6.4.1 Careful abstraction in memory—based learning
6.4.2 Getting started with FAMBL
6.4.3 Experiments with FAMBL
6.5 Conclusion
6.6 Further reading
7 Extensions
7.1 Wrapped progressive sampling
7.1.1 The wrapped progressive sampling algorithm
7.1.2 Getting started with wrapped progressive sampling
7.1.3 Wrapped progressive sampling results
7.2 Optimizing output sequences
7.2.1 Stacking
7.2.2 Predicting class n—grams
7.2.3 Combining stacking and class n—grams
7.2.4 Summary
7.3 Conclusion
7.4 Further reading
Bibliography
Index
· · · · · · (收起)
The approach is based on the combination of two powerful techniques: the efficient storage of solved examples of the problem, and similarity-based reasoning on the basis of these stored examples to solve new ones. (查看原文)
In general, probabilistic approaches have a greater coverage of syntactic constructions and vocabulary, they are more robust (they exhibit graceful degradation in deteriorating circumstances), they are reusable for different languages and domains, and development times for making applications and systems are shorter.
On the other hand, knowledge-based methods make the incorporation of linguistic knowledge and sophistication possible, allowing them to be more precise sometimes.
Probabilistic empirical methods are not without their own problems:
• The sparse data problem: often, not enough data is available to estimate the probability of events accurately. Some form of smoothing of the probability distributions is necessary.
• The relevance problem: it is often difficult to decide reliab... (查看原文)
还没人写过短评呢