The ApposCorpus: a new multilingual, multi-domain dataset for factual appositive generation

Proceedings of the 28th International Conference on Computational Linguistics, pages 1989–2003 Barcelona, Spain (Online), December 8-13, 2020 1989 The ApposCorpus: A new multilingual, multi-domain dataset for factual appositive generation Yova Kementchedjhieva University of Copenhagen yova@di.ku.dk Di Lu Dataminr, Inc. dlu@dataminr.com Joel Tetreault Dataminr, Inc. jtetreault@dataminr.com Abstract News articles, image captions, product reviews and many other texts mention people and orga- nizations whose name recognition could vary for different audiences. In such cases, background information about the named entities could be provided in the form of an appositive noun phrase, either written by a human or generated automatically. We expand on the previous work in ap- positive generation with a new, more realistic, end-to-end definition of the task, instantiated by a dataset that spans four languages (English, Spanish, German and Polish), two entity types (person and organization) and two domains (Wikipedia and News). We carry out an extensive analysis of the data and the task, pointing to the various modeling challenges it poses. The results we obtain with standard language generation methods show that the task is indeed non-trivial, and leaves plenty of room for improvement. 1 Introduction News articles, image captions, product reviews and many other texts mention people and organizations, whose name recognition could vary for different audiences. A piece of news, for example, may concern people and organizations that are known locally, but are not necessarily well-recognized on a global level. In such cases, news pieces targeted at a wider audience would provide background information about the entity in focus, often in the form of an appositive. For example: In March 2017 , Natalie Jaresko, former Minister of Finance in Ukraine, was appointed as the board’s executive director. It is unlikely that many people outside of Ukraine know the name Natalie Jaresko, so a foreign reader would likely benefit from the extra bit of information about her former occupation as justification for her new appointment. An appositive could also be less contextualized and provide information of more general importance, for example: The conservation unit is in the Calhau bairro of S˜ao Lu´ıs, the state capital. In general terms, appositives are phrases that appear next to a noun phrase and serve an explicative function (Bauer, 2017). Adding such explanations to text is a multi-step process. First, one has to decide whether an entity mention needs an appositive. That may not be the case for entities that are sufficiently well-known or that have been introduced earlier in the text. In case an appositive is indeed needed, the next step is to choose what information about the entity to disclose. If the information is to be of a factual nature, the writer needs to have prior knowledge of the entity, or access to an external knowledge resource–Kang et al. (2019) found appositives to be frequently based on facts of particular relevance to the context of the mention. Lastly, the surface form of the appositive, well-fitted to the surrounding context, needs to be produced. Viewed from the perspective of NLP, appositive generation …

Content truncated to 500 words