The SenSem (acronym for Sentence Semantics) Corpus for Catalan has been created as part of the SenSem Databank. Work on the Catalan corpus started in 2008 and continues in 2014. The Catalan Corpus was built by translating the journalistic subcorpus of the Spanish SenSem. It has received specific funding from the Spanish Government (Ministerio de Educación y Ciencia HUM2007-65267). The translation of the sentences has been carried out automatically and has been post-edited manually.

All the sentences have been annotated with syntactic and semantic information, resulting in a Catalan corpus of over 700,000 words, out of which 391,267 are annotated. These sentences exemplify almost 1,300 different verbs.

In the phrasal and sentence level, the different participants have been identified (differentiating between arguments and adjuncts). The most relevant syntactic-semantic information (semantic roles, syntactic functions and syntactic categories) has also been codified. The annotation of the internal structure of phrases has not been carried out.

Each sentence has also been associated to a constructional meaning and has had its formal mechanism described according to Goldberg’s (1995) concept of construction. This is, precisely, one of the novelties of the SenSem Corpus with respect to other resources of the same or similar type. The inclusion of the constructional information is key to completing the description of subcategorization patterns in order to avoid their ambiguity.

In addition to this, sentence aspectuality, polarity and modality have also been codified. These constitute valuable sources of information for the field of natural language processing (NLP). To start with, the lexical aspect (or Aktionsart) of verbs and phrases of sentences is useful in order to obtain the concatenation of events in the discourse, which is of interest in the question-answer interface. Similarly, differentiating between events and states is very useful, as well. In automatic generation, for example, the appropriate selection of a lexical item in some languages depends, precisely, on this aspectual information. On the other hand, polarity and modality are two key elements in the interpretation of factuality. In the field of NLP, the distinction between factive and non-factive events is crucial in order to tell which events described in texts are real and which are not (either because they have not taken place or because it is not know whether they have).

The annotation of the Catalan SenSem Corpus has been inhereted from its analogous Spanish project in an 85%. Sentences that present a different structure and/or function changes have been reannotated. These changes only affect the level of sentence complements and never their semantics. Due to this, the information about aspectuality, modality and polarity is strictly the same in both languages. The same applies to the semantics of constructions.

A corpus of a language such as Catalan, which does not usually have many available resources, with such diversified information and high-level semantics is of great interest in another, highly active, NLP field: the acquisition of information in order to create resources such as grammars and analyzers of different types.

One can search the Corpus online through the search engine available on the website: http://grial.uab.es/sensem/corpus. On this website, more specific documentation can be accessed, such as the definition of the terms used and their equivalence in other projects. All the publications related to this project can be consulted at http://grial.uab.es/publicacions.php.