Search

Computational Stylometry develops techniques that allow scholars to find out information about authors of texts by means of an automatic stylistic analysis. Indeed, each author’s style is unique, and no two authors are characterized by the same set of stylistic features. Several scholars focus on the analysis of different stylistic features and specific linguistic phenomena. The complex semantic aspects of figurative language remain one of the less investigated stylistic features, due to the lack of precise mapping between linguistic realizations and the concepts underlying the figurative language. Metaphors, similes, metonymies, and figurative language as a whole play an important role in how we mold our day-to-day reality, and they are represented through specific linguistic phenomena suitable to expressing specific conceptualization. Several linguistic devices and markers may highlight the presence of metaphorical expressions and other figures of speech. In this paper, we present research that exploits NooJ software and its linguistic resources to identify metaphors, oxymorons, and similes using a corpus of Italian dream descriptions, built by collecting dream descriptions from several blog users. Our goals, using NooJ’s features, are to build a domain dictionary able to represent lexical and semantic characteristics of metaphorical expressions, and to build model grammars able to recognize, extract, and tag these metaphors and figures of speech on the basis of different linguistic devices.

In this paper, we present the Archaeo-Term Project, along with one of its first efforts in enhancing multilingual access to Archaeological data, making available a resource of Archaeological terms within the framework of YourTerm CULT project. In order to enhance and promote the use of a terminological common ground across different languages the Archaeo-Term multilingual Glossary is intended both for scholars, experts in the field, translators and the general public. Its first release contains terms in Italian, English, German, Spanish and Dutch together with PoS, definitions and other linguistic information. This paper presents the data and the methodology adopted to create the glossary as well as the evaluation of the first results.

No matter what the main language of texts in the medical domain is, there is always an evidence of the usage of Latin-derived words and formative elements in terminology development. Generally speaking, this usage presents language-specific morpho-semantic behaviors in forming both technical-scientific and common-usage words. Nevertheless, this usage of Latin in Croatian medical texts does not seem consistent due to the fact that different mechanisms of word formation may be applied to the same term. In our pursuit to map all the different occurrences of the same concept to only one, we propose a model designed within NooJ and based on dictionaries and morphological grammars. Starting from the manual detection of nouns and their variations, we recognize some word formation mechanisms and develop grammars suitable to recognize Latinisms and Croatinized Latin medical terminology.

We describe two systems for semantic relation classification with which we participated in the SemEval 2018 Task 7, subtask 1 on semantic relation classification: an SVM model and a CNN model. Both models combine dense pretrained word2vec features and hancrafted sparse features. For training the models, we combine the two datasets provided for the subtasks in order to balance the under-represented classes. The SVM model performed better than CNN, achieving a F1-macro score of 69.98% on subtask 1.1 and 75.69% on subtask 1.2. The system ranked 7th on among 28 submissions on subtask 1.1 and 7th among 20 submissions on subtask 1.2.