Czech-English Bilingual Valency Lexicon Online

We describe CzEngVallex, a bilingual Czech–English valency lexicon which aligns verbal valency frames and their arguments. It is based on a parallel Czech-English corpus, the Prague Czech-English Dependency Treebank (PCEDT), where for each occurrence of a verb, a reference to the underlying Czech and English valency lexicons (PDT-Vallex and CzEngVallex, respectively) is recorded. The CzEngVallex then pairs the entries (verb senses) of the two lexicons, and allows for detailed studies of verb valency and argument structure in translation and also compare the approaches to valency in the two languages on the background of the same underlying theory, the Functional Generative Description. The CzEngVallex lexicon is now accessible online, and we will also describe here the search interface which makes certain complex queries possible, using the lexicon and accessing the associated examples of verb sense translations, as extracted from the PCEDT corpus. 1 The PCEDT parallel corpus and its lexicons Valency, or verb argument structure, is an important phenomenon both in linguistic studies as well as in language technology applications, since the verb is considered the core of a clause in (almost) every natural language utterance. Various dictionaries have been built from Propbank [13] to Framenet [1] as well as various valency lexicons exist for several languages, such as Walenty [16] for Polish, Verbalex [8] or Vallex [9] for Czech, Valence Lexicon for a Treebank of German [3] for German etc. However, there are no truly multilingual valency dictionaries linked to corpora. The Prague Czech-English Dependency Treebank (PCEDT 2.0) [4] contains the WSJ part of the Penn Treebank [10] and its manual professional translation to Czech, annotated manually using the tectogrammatical representation [11], first used for the Prague Dependency Treebank 2.0 (PDT) [5]. The tectogrammatical representation is in turn based on the Functional Generative Description theory [17].