Verbose, Laconic or Just Right: A Simple Computational Model of Content Appropriateness under Length Constraints

43478

Authors

nan

Abstract

Length constraints impose implicit re- quirements on the type of content that can be included in a text. Here we pro- pose the first model to computationally as- sess if a text deviates from these require- ments. Specifically, our model predicts the appropriate length for texts based on content types present in a snippet of con- stant length. We consider a range of fea- tures to approximate content type, includ- ing syntactic phrasing, constituent com- pression probability, presence of named entities, sentence specificity and inter- sentence continuity. Weights for these fea- tures are learned using a corpus of sum- maries written by experts and on high quality journalistic writing. During test time, the difference between actual and predicted length allows us to quantify text verbosity. We use data from manual eval- uation of summarization systems to as- sess the verbosity scores produced by our model. We show that the automatic ver- bosity scores are significantly negatively correlated with manual content quality scores given to the summaries. 1

Keywords

nan

Full Content

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 636–644, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Verbose, Laconic or Just Right: A Simple Computational Model of Content Appropriateness under Length Constraints Annie Louis∗ School of Infomatics University of Edinburgh Edinburgh EH8 9AB alouis@inf.ed.ac.uk Ani Nenkova Computer and Information Science University of Pennsylvania Philadelphia, PA 19103 nenkova@seas.upenn.edu Abstract Length constraints impose implicit re- quirements on the type of content that can be included in a text. Here we pro- pose the first model to computationally as- sess if a text deviates from these require- ments. Specifically, our model predicts the appropriate length for texts based on content types present in a snippet of con- stant length. We consider a range of fea- tures to approximate content type, includ- ing syntactic phrasing, constituent com- pression probability, presence of named entities, sentence specificity and inter- sentence continuity. Weights for these fea- tures are learned using a corpus of sum- maries written by experts and on high quality journalistic writing. During test time, the difference between actual and predicted length allows us to quantify text verbosity. We use data from manual eval- uation of summarization systems to as- sess the verbosity scores produced by our model. We show that the automatic ver- bosity scores are significantly negatively correlated with manual content quality scores given to the summaries. 1 Introduction In dialog, the appropriate length of a speaker turn and the amount of detail in it are hugely influ- enced by the pragmatic context. For example what constitutes an appropriate answer to the question “How was your vacation?” would be very different when the question is asked as two acquaintances pass each other in the corridor or right after two friends have ordered dinner at a restaurant. Simi- larly in writing, content is tailored to explicitly de- fined or implicitly inferred constraints on the ap- ∗Work done while at University of Pennsylvania. 50 word summary: The De Beers cartel has kept the diamond market stable by matching supply to demand. African nations have recently demanded better terms from the cartel. After the Soviet breakup, De Beers contracted for diamonds with the Yukutian Republic. The US remains the largest diamond market, followed by Japan. 100 word summary: The De Beers cartel, controlled by the Oppenheimer family controls 80% of the uncut diamond market through its Central Selling Organization. The cartel has kept the diamond market stable by maintaining a buffer pool of diamonds for matching supply to demand. De Beers opened a new mine in 1992 and extended the life of two others through underground mining. Innovations have included automated processing and bussing workers in daily from their homes. African nations have recently demanded better terms. After the Soviet breakup, De Beers contracted for diamonds with the Yukutian Republic. The US remains the largest diamond market, followed by Japan. Table 1: 50 and 100 word summaries written by the same person for the same set …

Content truncated to 500 words

Labeled Text

Length constraints impose implicit requirements on the type of content that can be included in a text. Here we propose the first model to computationally assess if a text deviates from these requirements. Specifically, our model predicts the appropriate length for texts based on content types present in a snippet of constant length. We consider a range of features to approximate content type, including syntactic phrasing, constituent compression probability, presence of named entities, sentence specificity and intersentence continuity. Weights for these features are learned using a corpus of summaries written by experts and on high quality journalistic writing. During test time, the difference between actual and predicted length allows us to quantify text verbosity. We use data from manual evaluation of summarization systems to assess the verbosity scores produced by our model. We show that the automatic verbosity scores are significantly negatively correlated with manual content quality scores given to the summaries. 1

Publication Details

Paper ID:
43478
Year:
2014
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan