What's in?

Table of contents

• Corpus composition
• Companion tools and online services
• Annotation guidelines
• Tagset
  • Part-of-speech tags
  • Inflection tags
  • Named-entity tags

Corpus composition

CINTIL is a corpus of Portuguese with 1 Million annotated tokens, each one of which verified by human expert annotators. The annotation comprises information on part-of-speech, open classes lemma and inflection, multi-word expressions pertaining to the class of adverbs and to the closed POS classes, and multi-word proper names (for named entity recognition).

Over one third of the corpus is composed of transcribed spoken materials, with about half of that being the transcription of informal conversations.

The remaining corpus is composed of written materials. The majority (58.73%) of this written corpus includes articles from newspapers and magazines, such as the Jornal Público, Diário de Notícias, Revista Visão, etc. The rest of the written corpus is mostly composed of fiction.

A more detailed breakdown of the composition of CINTIL is presented in the following table:

Companion tools and online services

They include the following individual tools covering analysis and generation procedures:

• Sentence chunker: detects and marks paragraph and sentence boundaries; 99.94% accuracy
• Tokenizer: segments text into tokens, expands contractions, detaches clitic pronouns from verbs, etc.; 99.72% accuracy
• POS tagger: assigns POS tags to tokens in context; 96.87% accuracy
• Nominal featurizer: assigns inflection features (gender and number; and person and degree when applicable) to (attested or putative) words from the nominal POS categories, resolving ambiguity in context; 91.07% f-score
• Nominal lemmatizer: assigns a lemma to words from the nominal POS categories (viz. common nouns and adjectives), resolving ambiguity in context; 97.67% f-score
• Verbal featurizer and lemmatizer: assigns inflection features (tense, person and number) and lemma (infinitive form) to any (attested or putative) verb form, resolving ambiguity in context; 95.96% f-score
• Verbal conjugator: delivers a conjugation table given any (attested or putative) infinitive verb form and the specification of possible associated clitics
• Nominal inflector: delivers an inflected form given another (attested or putative) form and the inflected features required for the output

• LX-Inflector (http://lxinflector.di.fc.ul.pt) This service allows to obtain the lemma and the inflected form from any nominal form provided (common nouns or adjectives - neologisms included), and according to a given specification of inflection feature values entered. This functionality is supported by the composition of the nominal featurizer, lemmatizer and inflector tools indicated above.
• LX-Lemmatizer (http://lxlemmatizer.di.fc.ul.pt) This service provides fully-fledged lemmatization of Portuguese verbs: given a verb form (neologisms included), it delivers all possible lemmata together with their respective inflection features.
• LX-Conjugator, (http://lxconjugator.di.fc.ul.pt) This service provides fully-fledged conjugation tables for any verb form entered (neologism included), including the full range of pronominal forms.
• LX-Suite (http://lxsuite.di.fc.ul.pt) This service is supported by the composition of the range of processing tools for analysis: sentence chunker, tokenizer, POS tagger, nominal featurizer, nominal lemmatizer and verbal lemmatizer. The resulting functionality ensures that the input, entered as raw text, is sentence and token-segmented, their tokens are associated with corresponding lemmata and tagged with linguistic information on their POS and inflection feature values. The outcome is resolved with respect to ambiguity that arise at the different levels of processing.

Annotation guidelines

Tagset

Part-of-speech tags

Inflection tags

Named-entity tags