Automatic Analysis of Semantic Coherence in Academic Abstracts Written in Portuguese

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1144–1152, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Automatic Analysis of Semantic Coherence in Academic Abstracts Written in Portuguese Vin´ıcius Mour˜ao Alves de Souza State University of Maring´a Maring´a, PR, Brazil, 87020-900 vsouza@din.uem.br Val´eria Delisandra Feltrim State University of Maring´a Maring´a, PR, Brazil, 87020-900 valeria.feltrim@din.uem.br Abstract SciPo is a system whose ultimate goal is to support novice writers in produc- ing academic texts in Brazilian Portuguese through presentation of critiques and sug- gestions. Currently, it focuses on the rhetorical structure of texts, being capa- ble of automatically detecting and criti- cizing the rhetorical structure of Abstract sections. We describe a system that en- hances SciPo’s functionality by evaluating aspects of semantic coherence in academic abstracts. This system identifies features of sentences based on semantic similar- ity measures and rhetorical structure. Dif- ferent machine learning algorithms were trained and evaluated with these features, resulting in three classifiers capable of de- tecting specific coherence issues on sen- tences with regard to a rhetorical structure model for abstracts. Results indicate that the system yields higher performance than the baseline for all classifiers. 1 Introduction This research has been motivated by a need for advanced discourse analysis capabilities for writ- ing tools such as SciPo (short for Scientific Por- tuguese). SciPo (Feltrim et al., 2006) is a sys- tem whose ultimate goal is to support novice writ- ers in producing academic texts in Brazilian Por- tuguese. Currently, it focuses on Computer Sci- ence academic texts and supports the writing of abstracts and introductions. Its functionalities are based on the use of structure models — in terms of schematic structure, rhetorical strategies and lex- ical patterns — similar to the ones proposed by Swales (1990) and Weissberg and Buker (1990), and authentic examples organized as case bases. Although SciPo provides feedback with regard to the text rhetorical structure in the form of critiques and suggestions, it does not provide considerations about the text semantics, such as aspects related to its coherence, which is a fundamental characteris- tic for text legibility and interpretability. We understand coherence as what makes a group of words or sentences semantically mean- ingful. We assume that coherence refers to the establishment of a logical sense among different sentences of a text. Thus, it is a principle of in- terpretability related to the communicational sit- uation and to the capability of the reader in cal- culating the meaning of the text. Therefore, it is bounded to the text, but it does not depend only on the text (van Dijk, 1981). Aiming at complementing SciPo’s functionali- ties, we have developed classifiers for the auto- matic detection of specific semantic relations in academic texts in Portuguese, then it can be used by SciPo for providing feedback referring to text coherence. Based on textual features that can be readily read off the text, the classifiers present indications related to semantic aspects that con- tribute to a high level of coherence. …

Content truncated to 500 words