Tutorial 3: Textual Entailment
Ido Dagan, Dan Roth, Fabio Massimo Zanzotto
Recognizing Textual Entailment is the task of determining, for example, that the sentence: "Google files for its long awaited IPO" entails that "Google goes public". Determining whether the meaning of a given text passage entails that of another or whether they have the same meaning is a fundamental problem in natural language understanding that requires the ability to abstract over the inherent syntactic and semantic variability in natural language. This challenge is at the heart of many natural language understanding tasks including Question Answering, Information Retrieval and Extraction, Machine Translation, and others that attempt to reason about and capture the meaning of linguistic expressions. The task has attracted significant interest over the last couple of years mainly fostered by the PASCAL Recognizing Textual Entailment Challenge (RTE). A substantial number of papers on these topics have been published in major conferences and workshops in the last couple of years.
The primary goals of this tutorial are to review the framework of applied Textual Entailment and motivate it as a generic paradigm for natural language semantics. We will present some of the key computational approaches proposed and some of the obstacles identified by the research community in this area, as a way to promote further research. The tutorial will thus be useful for many of the senior and junior researchers that have prior or new interest in this area, providing a concise overview of recent perspectives and research results.
Tutorial Outline
-
Motivation and Task Definition
- Textual Entailment as a generic (application independent) semantic inference test.
- Motivation.
- Textual Entailment as a framework for the study of applied semantics.
-
A Skeletal review of Textual Entailment Systems
- A systematic approach to textual entailment: an architecture.
- A survey of existing approaches: a unified perspective.
- Knowledge representation and knowledge resources.
- Decision processes based on inference paradigms.
- Decision processes based on machine learning.
-
Knowledge Acquisition Methods
- Textual entailment use of linguistic and world knowledge.
- Knowledge as augmenting the surface level text representation.
- Knowledge base support of textual entailment.
- Knowledge acquisition methods.
- The interaction between text representation and the knowledge base.
-
Applications of Textual Entailment
- The utility of the Textual Entailment setting for applications.
- Proposing ways to use generic entailment models in specific applications.
-
A Textual Entailment view of Semantics
- Textual Entailment as a vehicle for the study of old and new semantic phenomena.
- A unified evaluation schemes for semantic tasks
- Redefining semantic problems via Textual Entailment
Ido Dagan has broad research experience and publication record in various areas of empirical natural language processing. He has presented several conference tutorials and summer school courses. In particular, Ido is interested in applied semantic modeling, facilitated largely through unsupervised learning approaches. In the last few years Ido and his colleagues introduced textual entailment as a generic framework for applied semantic inference and have been the main organizers of the first three rounds of the PASCAL Recognizing Textual Entailment Challenges.
Dan Roth has published broadly in machine learning, natural language processing, knowledge representation and reasoning and received several best paper and research awards. He has developed several machine learning based natural language processing systems including an award winning semantic parser, and has presented invited talks in several international conferences, and several tutorials on machine learning for NLP. Over the last few years he has worked on machine learning and inference methods in the context of natural language understanding tasks and, in particular, on textual entailment. He has presented an invited talk on this topic at the workshop on Empirical Modeling of Semantic Equivalence and Entailment that was collocated with ACL-2005.
Fabio Massimo Zanzotto is an associate professor at the University of Rome "Tor Vergata". He has been working in building models for robust syntactic parsing and for knowledge acquisition from corpora. In the last three years, he worked on the definition of textual entailment recognition models. He mainly explored the application of supervised machine learning models to learn inference rules from annotated examples. He participated in the first and the second Pascal RTE challenges with two different systems that explore different models for the use of supervised learning algorithms to the Textual Entailment recognition problem.
