There are many branches of NLP research which involve the generation of language (summarisation, MT, human-computer dialogue, application front-ends, data-to-text generation, document authoring, etc.). However, it is not always easy to identify common ground among the generation components of these application areas, which has sometimes made it difficult for generic research in 'Natural Language Generation' (NLG) to engage with them effectively. Recent advances in corpus-based approaches (both manual and automatic) across many of these areas, and in particular in NLG itself, offer a new perspective on this problem and the opportunity to explore synergies and differences from the secure common grounding of corpus data.

This workshop is the third in an occasional series seeking to exploit this opportunity by providing a forum for discussing NLG and its links with these closely related fields from a corpus-oriented perspective. These workshops have the general aims:

• to provide a forum for reporting and discussing corpus-oriented methods for generating language;
• to foster cross-fertilisation between NLG and related fields by looking for common ground through corpus-oriented approaches;
• to promote the sharing of data and methods in all language generation research.

There are two basic approaches to text summarisation: abstractive, where texts are analysed, the internal representations are pruned, and a more condensed version regenerated, and extractive, where key passages of the input texts themselves are identified and the 'glued together' to form a shorter text. Extractive summarisation is less dependent on fragile 'deep' analysis and regeneration techniques, but tends to produce summaries that are not very coherent and whose referring expressions are not very clear (so for example, they often score low on DUC human assessment criteria such as 'discourse coherence' and 'referential clarity').

The relevance of NLG techniques to abstractive summarisation is clear, but recently there has also been increasing interest in regeneration as a post-process for extractive summaries. Work by Otterbacher et al., Steinberger et al. and Nenkova et al., for example, show how regeneration of (parts of) extractive summaries can increase their coherence, referential clarity or fluency. At the same time NLG researchers are investigating techniques that could be used to improve extractive summaries by regenerating them (in particular in the subfield of referring expression generation, see for example the GREC Task papers at INLG 2008).

The core aim of this workshop is to provide a forum for NLG and summarisation researchers to examine the similarities and differences between their current approaches to generating language, and to explore the potential for cross-fertilisation. To this end, the workshop will include:

• a keynote by Professor Kathy McKeown (Columbia) - "Query-focused Summarization Using Text-to-Text Generation: When Information Comes from Multilingual Sources" ;
• a panel-led discussion session on links between NLG and summarisation; and
• a session on shared tasks of interest to researchers in both communities.

This will be supported by a programme of technical papers, on all aspects of using corpora in the generation of language, with a particular interest in relevance to text summarisation. Specific topics include, but are not limited to:

• generation techniques in abstractive summarisation
• regeneration/rewriting/post-processing techniques for extractive summarisation
• generation of references to named entities in discourse context
• annotating corpora for language generation and summarisation
• uses of corpora in the evaluation of language generation and summarisation systems
• reuse of corpus resources developed for NLU (e.g. treebanks) in language generation and summarisation
• domain-specific vs. general-purpose corpora for language generation and summarisation
• statistical approaches to language generation
• machine learning methods for language generation

Invited speaker:

Panelists:

Organisers:

Programme committee:

Accepted papers

Keynote paper:

Kathy McKeown
Query-focused Summarization Using Text-to-Text Generation: When Information Comes from Multilingual Sources

Hide abstract

Abstract

The past five years have seen the emergence of robust, scalable natural language processing systems that can summarize and answer questions about online material. One key to the success of such systems is that they re-use text that appeared in the documents rather than generating new sentences from scratch. Re-using text is absolutely essential for the development of robust systems; full semantic interpretation of unrestricted text is beyond the state of the art. Better summaries and answers can be produced, however, if systems can generate new sentences from the input text, fusing relevant phrases and discarding irrelevant ones. When the underlying sources for question answering come from multiple languages, the need for text-to-text generation is even more pronounced.

This talk will present research on query-focused summarization over a variety of sources, including news, broadcast news, talks shows and blogs. Our research combines approaches from summarization and information extraction to answer open-ended questions. Because our sources include informal genres as well as formal genres and draw from English, Arabic and Chinese, text-to-text generation is critical for improving the intelligibility of responses. In this talk I will describe how we exploit information available at question answering time to edit sentences, removing redundant and irrelevant information and correcting errors in translated sentences.

Long papers:

Karolina Owkzarzak and Hoa Trang Dang
Evaluation of automatic summaries: Metrics under varying data conditions

Horacio Saggion
A Classification Algorithm for Predicting the Structure of Summaries

Jackie Chi Kit Cheung, Giuseppe Carenini and Raymond Ng
Optimization-based Content Selection for Opinion Summarization

Wei Xu and Ralph Grishman
A Parse-and-Trim Approach with Information Significance for Chinese Sentence Compression

Hideki Tanaka, Akinori Kinoshita, Takeshi Kobayakawa, Tadashi Kumano and Naoto Katoh
Syntax-Driven Sentence Revision for Broadcast News Summarization

João Cordeiro, Gaël Dias and Pavel Brazdil
Unsupervised Induction of Sentence Compression Rules

Short Papers:

Stephanie Schuldes, Michael Roth, Anette Frank and Michael Strube
Creating an Annotated Corpus for Generating Walking Directions

Iris Hendrickx, Walter Daelemans, Erwin Marsi and Emiel Krahmer
Reducing Redundancy in multi-document Summarization Using Lexical Semantic Similarity

Maria Fernanda Caropreso, Diana Inkpen, Shahzad Khan and Fazel Keshtkar
Visual Development Process for Automatic Generation of Digital Games Narrative Content

Mohit Kumar, Dipanjan Das, Sachin Agarwal and Alexander Rudnicky
Non-textual Event Summarization by Applying Machine Learning to Template-based Language Generation

Last modified: 22 Jul 2009
Maintained by: