Extraction of Natural-Language Dates and Comparison of Dates in Hypothesis and Text to Identify Negative Textual Entailment
Public Deposited- Last Modified
- February 28, 2019
- Creator
-
Orphanides, Andreas K.
- Affiliation: School of Information and Library Science
- Abstract
- We present a system to determine entailment, when a sentence (the text) implies a second sentence (the hypothesis). While some systems use temporal information to decide entailment, no study has measured the effectiveness of temporal features alone in entailment resolution. The system identifies natural-language dates, and precludes entailment solely by comparing dates in the hypothesis and text. Evaluation of date detection on three 800-pair corpora from the Recognising Textual Entailment (RTE) Challenges provided precision of 98.0% (RTE3devmt, 338/345), 96.6% (RTE3test, 198/205), and 99.7% (RTE1test, 336/337), and recall of 97.7% (RTE3devmt, 338/346), 99.0% (RTE3test, 198/200), and 99.1% (RTE1devmt 336/339). For sentence pairs with years, the proposed method improved entailment accuracy from 33 to 42/72 (RTE3devmt) and from 42 to 44/63 (RTE3test,) which corresponds to an overall improvement of 1.1% (RTE3devmt) and 0.1% (RTE3test). Our analysis suggests that matching temporal information with an event would further increase the entailment accuracy.
- Date of publication
- April 2008
- Subject
- DOI
- Resource type
- Rights statement
- In Copyright
- Advisor
- Blake, Catherine L.
- Degree granting institution
- University of North Carolina at Chapel Hill
- Language
- Extent
- 66 p.
- Access right
- Open access
- Date uploaded
- December 10, 2010
Relations
- Parents:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Extraction of Natural-Language Dates and Comparison of Dates in Hypothesis and Text to Identify Negative Textual Entailment. | 2019-05-13 | Public | Download |