Wka project

Course Project: Temporal relation extraction

Many NLP applications such as text summarization, question-answering, event tracking, and maybe ontology extraction depend on, or can greatly benefit from modeling time properly within the context of a document. The 3 types of temporal relations, according to the SemEval 2007 (see TimeEval ref below) shared task are:

• Task A: between an event and a time expression within the same sentence
• Task B: between an event and the document creation time
• Task C: between a pair of events in adjacent sentences

We want to experiment with existing extraction methods and attempt to add new/improve existing methods.

Data

• TimeBank. http://www.timeml.org/site/timebank/timebank.html
  • Few hundreds of documents annotated with event and time-related tags in TimeML
  • Annotations are sparse and defficient in this dataset, and the corpus is known for relatively low inter-annotator agreement.
    • [Chambers and Jurafsky] address this problem using transitive closures to expand the training data, and by using annotations from [Bethard et al.].
• Others might be available

Plan

Work on this project will take 3 directions:

• Expansion of the (small) dataset
• Features to be extracted and used: verb attributes, time expressions, syntax trees, etc
  • Along with tools needed: POS tagging, parsing, etc
• Learning methods
  • will depend on feasibility within project time constraints and learning achieved in first period

Approaches

Various methods have been used for learning in the literature

• Markov Logic Networks
• Integer Linear Programming
• Syntactic-Semantic Analysis
• Maximum Entropy, SVMs, etc

Team

One person (user:wka).

References

• [1] Verhagen et al. SemEval-2007 Task 15: TempEval Temporal Relation Identification.
• [2] Yoshikawa et al. Jointly identifying temporal relations with Markov logic.
• [3] Chambers and Jurafsky. Jointly combining implicit constraints improves temporal ordering.
• [4] I. Mani. Recent developments in temporal information extraction.
• [5] G. Puscasu. WVALI: Temporal relation identification by syntactico-semantic analysis.
• [6] Bethard and Martin. CU-TMP: Temporal relation classification using syntactic and semantic features.
• [7] Mani et al. Machine learning of temporal relations.
• [8] Richardson and Domingos. Markov logic networks.
• [9] J.F. Allen, 1993. Maintaining knowledge about temporal intervals.
• [10] Allen and Hayes. A common-sense theory of time.
• [11] I. Mani and G. Wilson. Robust temporal processing of news.
• [12] Hwang and Schubert, 1992. Tense trees as the "fine Structure" of discourse.
• [13] Bethard et al. Timelines from text: identification of syntactic temporal relations.