1 / 10

Natural Language Processing for the Web

Natural Language Processing for the Web. Prof. Kathleen McKeown 722 CEPSR, 939-7118 Office Hours: Wed, 1-2; Tues 4-5 TA: Yves Petinot 719 CEPSR, 939-7116 Office Hours: Thurs 12-1, 8-9. Projects. Interim results due next Thursday (3/11) Hand in via courseworks (by midnight )

janet
Download Presentation

Natural Language Processing for the Web

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Natural Language Processing for the Web Prof. Kathleen McKeown 722 CEPSR, 939-7118 Office Hours: Wed, 1-2; Tues 4-5 TA: Yves Petinot 719 CEPSR, 939-7116 Office Hours: Thurs 12-1, 8-9

  2. Projects • Interim results due next Thursday (3/11) • Hand in via courseworks (by midnight) • Guest Lecturer next Thursday: Regina Barzilay, MIT • We will meet in the Interschool Lab, 7th Floor CEPSR (by my office)

  3. Today • All papers are on the same problem • Can we rank them considering: • Presentation • Originality • Results

  4. Learning to Recognize Features of Valid Entailment • Typical approach • Aligns sentence with hypothesis • Produce all possible alignments and return the best scoring one • Closeness of match determines entailment • Problems with this approach • Assumption of monoticity • Assumption of locality • Confounding of alignment with entailment evaluation

  5. Possible alignments • Sharon warns Arafat could be targeted for assassination • The prime minister targeted for assassination • Space of alignments: O((m+1)n) for a sentence of m nodes and hypothesis of n

  6. Word overlap: 2081 a problem • Upward monoticity problem: 98 a problem • Sharon denies Arafat targeted for assassination • Global phenomena: Dogs barked loudly -> dogs barked. No dogs barked loudly does not • What about 231?

  7. Solution • Alignment first • Between dependency parses • Using Wordnet to identify synonyms and collations plus McCallum’s NE tagger • Don’t worry about quantification, modalities • Machine learning for binary classification • Takes into account context in which alignment occurs

  8. Entailment Classification • Logistic regression • 28 features • Polarity: negation, downward monotone quantifiers (few, no), restricting preps (without, except), superlatives (tallest) • Adjuncts: (loudly) • Antonymy: (rose vs fell) • Modality: • map text to modality: possible, not possible, actual, not actual, necessary, not necessary. • Modality pair: yes, weak yes, don’t know, weak no, no • (not possible -> not actual)? -> yes • (possible -> necessary) -> weak no

  9. Factivity: (tried to, managed to) • Quantifiers: no, some many, most, all • Number,date and time features (can do fuzzy matches) • Alignment features: a good score or not? • Best features: adjunct, dates, modals

More Related