NOMIT: Automatic Titling by Nominalizing

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 274–283, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics NOMIT: Automatic Titling by Nominalizing Cédric Lopez, Violaine Prince, and Mathieu Roche LIRMM, CNRS, Univ. Montpellier 2 161, rue Ada Montpellier, France {lopez,prince,mroche}@lirmm.fr Abstract The important mass of textual documents is in perpetual growth and requires strong ap- plications to automatically process informa- tion. Automatic titling is an essential task for several applications: ’No Subject’ e-mails ti- tling, text generation, summarization, and so forth. This study presents an original ap- proach consisting in titling journalistic articles by nominalizing. In particular, morphological and semantic processing are employed to ob- tain a nominalized form which has to respect titles characteristics (in particular, relevance and catchiness). The evaluation of the ap- proach, described in the paper, indicates that titles stemming from this method are informa- tive and/or catchy. 1 Introduction A title establishes a link between a reader and a text. It has two main functions. First of all, a ti- tle can be informative (it conveys relevant informa- tion about the text content and aim), and second, it can be catchy or incentive (Herrero Cecilia, 2007). A heading is said to be catchy when it succeeds in capturing the reader’s attention on an aspect of the announced event, in a ingenious, metaphoric, enig- matic, or shocking way. From a syntactic point of view, a title can be a word, a phrase, an expression, a sentence, that designates a paper or one of its parts, by giving its subject. Titles are used within applications such as auto- matic generation of contents, or summarization. So, it is interesting to automate the process that produces relevant titles by extracting them from texts, and supplying other applications with such data, while avoiding any human intervention: Direct applica- tions (as automatic titling of "no object" e-mails) are thus possible. The point is that several titles can be relevant for a same text: This constitutes the main difficulty of au- tomatic titling. Some writers prefer informative ti- tles, whereas others prefer catchy ones. Others jug- gle with both criteria according to the context and the type of the publication. So, evaluation of au- tomatic titling is a complex step requiring a human intervention. Indeed, how can titles relevance be es- timated ? How an automatic title can be compared to a human-written ("real") title, knowing that both can have a very different morphosyntactic structure? Automatic titling is a full process, possessing its own functions. It has to be sharply differentiated from summarization and indexation tasks. Its pur- pose is to propose title(s) that have to be short, infor- mative and/or catchy, and keep a coherent syntactic structure. NLP1 methods will be exploited in order to abide by language morphosyntactic and semantic constraints in titling. In this paper, we describe an approach of auto- matic titling relying on nominalization, i.e. rules transforming a verb phrase into a noun phrase (e.g. "the …

Content truncated to 500 words