Page 1
Combining Lexical Semantic Resources with Question & Answer Archives for Translation-Based Answer Finding
Delphine Bernhard and Iryna GurevvchUbiquitous Knowledge Processing (UKP) Lab
Computer Science DepartmentTechnische Universit¨at Darmstadt, Hochschulstraße 10
D-64289 Darmstadt, Germany
ACL 2009
Reporter: Kan-Wen TienDate: 2009.10.22
Page 2
Outlines
• Introduction• Related Work• Parallel Datasets• Semantic Relatedness Experiments• Answer Finding Experiments• Conclusion
Page 3
• Introduction• Related Work• Parallel Datasets• Semantic Relatedness Experiments• Answer Finding Experiments• Conclusion
Page 4
Introduction
• Lexical gap between queries and documents or questions and answers
• Several solutions:– Query reformulation, query paraphrasing– Query expansion – Semantic information retrieval
Page 5
Introduction
• Several solutions:– Integrate monolingual statistical translation
models in the retrieval process (1999)• Drawback: limited availability of truly parallel
monolingual corpora
• Training data often consist in question-answer pairs and usually extracted from the evaluation corpus itself
Page 6
• Introduction
• Related Work• Parallel Datasets• Semantic Relatedness Experiments• Answer Finding Experiments• Conclusion
Page 7
Related Work
• Statistical translation models for retrieval• Built synthetic training data • Train translation models on Q&A pairs – Answers -> source language– Questions -> target language
• Select the most important terms to build compact translation models
Page 8
• Introduction• Related Work
• Parallel Datasets• Semantic Relatedness Experiments• Answer Finding Experiments• Conclusion
Page 9
Parallel Datasets
• Different data resources:(1)Manually-tagged question reformulations
and question-answer pairs from the WikiAnswers social Q&A site
(2) Glosses from WordNet, Wiktionary, Wikipedia and Simple Wikipedia
Page 11
Parallel Datasets
(1) Manually-tagged question reformulations and question-answer pairs
• From social Q&A sites: WikiAnswers (WA)– Question-Answer Pairs (WAQA)
– Question Reformulations (WAQ)
[URL]
Page 12
Parallel Datasets
(2) Glosses from WordNet, Wiktionary, Wikipedia and Simple Wikipedia
• Lexical Semantic Resources (LSR)– Word sense alignment
• Example !
Page 13
Parallel Datasets
• Example: “moon”– Wordnet (sense 1): The natural satellite of the
Earth.– English Wiktionary: The Moon, the satellite of
planet Earth.– English Wikipedia: The Moon (Latin: Luna) is
Earth’s only natural satellite and the fifth largest natural satellite in the Solar System.
Page 14
Parallel Datasets
Three datasets: • Question-Answer Pairs (WAQA)
1,227,362 parallel pairs
• Question Reformulations (WAQ)4,379,620 parallel pairs
• Lexical Semantic Resources (LSR)397,136 pairs
Page 15
Parallel Datasets
• Translation Model Training– Pre-processing steps
– GIZA++ SMT Toolkit -> word-to-word translation probabilities
– IBM translation model 1
Page 16
Parallel Datasets
• Combination of the datasets– Lin (combination of models after training)
– Pool (concatenating the corpora before training)
Page 17
Parallel Datasets
Page 18
Parallel Datasets
Page 19
Parallel Datasets
Page 20
• Introduction• Related Work• Parallel Datasets
• Semantic Relatedness Experiments• Answer Finding Experiments• Conclusion
Page 21
Semantic Relatedness Experiments• Goal: Word translation probabilities vs.
Concept vector based measure
• Concept vector based measure relying on Explicit Semantic Analysis(Gabrilovich and Markovitch, 2007)
• Compare with traditional semantic relatedness measures
Page 23
Semantic Relatedness Experiments
Page 24
Semantic Relatedness Experiments
• Testing data set: 353 word-to-word pairs– Created by Finkelstein et al. (2002)– Fin1-153: 153 pairs– Fin2-200: 200 pairs
Page 25
Semantic Relatedness Experiments
• Testing data set: 353 word-to-word pairs– Created by Finkelstein et al. (2002)– Fin1-153: 153 pairs– Fin2-200: 200 pairs
Page 26
Semantic Relatedness Experiments
• Use Spearman’s Rank Correlation Coefficients (-1, 0, +1)
[URL]
Page 27
Semantic Relatedness Experiments
• Use Spearman’s Rank Correlation Coefficients (-1, 0, +1)
[URL]
Page 28
• Introduction• Related Work• Parallel Datasets• Semantic Relatedness Experiments
• Answer Finding Experiments• Conclusion
Page 29
Answer Finding Experiments
• Goal: provide an extrinsic evaluation of the translation probabilities by employing them in an answer finding task.
• Using a ranking function to perform retrieval
Page 30
Answer Finding Experiments
• Ranking function (β = 0.8, λ = 0.5)
Page 31
Answer Finding Experiments
• Ranking function (β = 0.8, λ = 0.5)
Page 32
Answer Finding Experiments
• Ranking function (β = 0.8, λ = 0.5)
Query likelihood modelTranslation model
Page 33
Answer Finding Experiments
• Testing data: Microsoft Research QA Corpus• 1,364 questions, 9,780 answers• 5 levels of relevance judgements:
0: No Judgement Made1: Extract Answers3: Off Topic4: On Topic, Off Target5: Partial Answer
Page 34
Answer Finding Experiments
• Testing data: Microsoft Research QA Corpus• 1,364 questions, 9,780 answers• 5 levels of relevance judgements:
0: No Judgement Made1: Extract Answers3: Off Topic4: On Topic, Off Target5: Partial Answer
Page 35
Answer Finding Experiments
Page 36
Answer Finding Experiments
• Mean Average Precision (MAP)• Mean R-Precision (R-prec)• Baselines: – Query likelihood model (QLM) ---> β = 0
– LuceneQuery likelihood model Translation model
Page 37
Answer Finding Experiments
Page 38
Answer Finding Experiments
Page 39
Answer Finding Experiments
Page 40
Answer Finding Experiments
Page 41
Answer Finding Experiments
Page 45
• Introduction• Related Work• Parallel Datasets• Semantic Relatedness Experiments• Answer Finding Experiments
• Conclusion
Page 46
Conclusion
• Propose new kinds of datasets for training• Provide the first intrinsic evaluation of word
translation probabilities with respect to human relatedness rankings for reference word pairs
• Models based on translation probabilities for answer finding