CINTIL DependencyBank PREMIUM - A Corpus of Grammatical Dependencies for Portuguese

CINTIL DependencyBank PREMIUM A Corpus of Grammatical Dependencies for Portuguese Rita de Carvalho, Andreia Querido, Marisa Campos, Rita Valadas Pereira, Jo˜ao Silva, Ant´onio Branco University of Lisbon Departamento de Inform´atica, Faculdade de Ciˆencias, Universidade de Lisboa Campo Grande, 1749-016 Lisboa, Portugal {rita.carvalho, andreia.querido, doriela.campos, ana.pereira, jsilva, antonio.branco}@di.fc.ul.pt Abstract This paper presents a new linguistic resource for the study and computational processing of Portuguese. CINTIL DependencyBank PRE- MIUM is a corpus of Portuguese news text, accurately manually annotated with a wide range of linguistic information (morpho-syntax, named-entities, syntactic function and semantic roles), making it an invaluable resource specially for the development and evaluation of data-driven natural language processing tools. The corpus is under active development, reaching 4,000 sentences in its current version. The paper also reports on the training and evaluation of a dependency parser over this corpus. CINTIL DependencyBank PREMIUM is freely-available for research purposes through META-SHARE. Keywords: dependency bank, corpora, dependency parsing 1. Introduction Linguistically annotated data sets are of the utmost impor- tance for the study and computational processing of natural languages. In particular, they enable the training and eval- uation of data-driven NLP tools, such as taggers, chunkers and parsers. Over the past years, the focus of parsing tasks has moved from determining phrase constituency towards assigning syntactic function, also known as grammatical dependen- cies, as this sort of annotation further abstracts from the surface form of a language and is better suited for some tasks—particularly data extraction tasks—since it provides a direct representation of predicate-argument relations. As a result, data sets annotated with grammatical dependen- cies, or dependency banks, have also become popular (de Marneffe et al., 2006). This paper presents CINTIL DependencyBank PREMIUM, a corpus of Portuguese newspaper text manually annotated with syntactical and semantic dependency relations, as well as part-of-speech, inflection, lemmas and named entities, making it a new important resource for NLP and for the linguistic analysis of Portuguese. CINTIL DependencyBank PREMIUM is a new treebank, different from the existing top quality dependency bank, CINTIL-DepBank (Branco et al., 2011), in that the latter, although it uses the same tag set and annotation scheme, was built with the help of the LXGram computational gram- mar (Costa and Branco, 2010) and thus includes only those sentences that the grammar was able to parse, a fact that limits the coverage of that corpus to the coverage of its supporting grammar. The new CINTIL DependencyBank PREMIUM, in turn, does not have this limitation as it in- cludes all sentences in the raw base corpus, instantiating all kinds of grammatical phenomena and lexical items that actually occur in the corpus. CINTIL DependencyBank PREMIUM is actively devel- oped and maintained by NLX, the Natural Language and Speech Group of the Department of Informatics of the Uni- versity of Lisbon. It is available through META-SHARE. This paper is structured as follows. Section 2. introduces CINTIL DependencyBank PREMIUM, presenting figures about the corpus and the tag set. Section 3. describes the annotation methodology and Section 4. describes the lin- guistic phenomena …

Content truncated to 500 words