Repository logo

Browsing by Author "Kakanaeeni, Bahareh"

Now showing 1 - 1 of 1
  • An electronic dictionary of persian verbs
    Publication . Kakanaeeni, Bahareh; Baptista, Jorge
    There are more than 110 million Persian speakers in the world, but the lexical resources for Natural Language Processing (NLP) of the Persian language are still too scarce. Though Persian (or Fārsi) uses an adapted form of the Arabic script, it is an Indo-European language and its verbal inflection is based on stems and affixes. This project is to be considered as a first step towards the construction of large-scale lexical resources for Persian, and the development of a Persian module to be distributed with the Unitex linguistic development platform. The specific goal of this dissertation is to build a morphologic, machine-readable dictionary of Persian verbs, using a dictionary of lemmas and a set of morphologic finite-state transducers (FST) to generate all the inflected forms associated to each lemma, and encode them with all the relevant morphosyntactic information (tense, person-number, etc.). This task is complicated in Persian verbal morphology by the fact that each verb has two stems (past and present), different inflection paradigms are used for written (formal) and oral (informal) language uses, and several compound tenses can be formed through combining prefixes and suffixes with base inflected forms, the same tense being able to constitute one, two or more different tokens (separate written forms). A small dictionary of lemmas (145) and their respective written (292) and spoken (127) stems was built, each stem was provided with the appropriate inflection conventional code, which correspond to an inflection paradigm. The list of lemmas was compiled based on frequency data from a large Persian corpus, the TEP (Tehran English-Persian Parallel Corpus), containing around 4.5 million words, and selecting the most frequent verb forms. At its current state, the dictionary of inflected forms contains 1,536 entries. In Persian, in average, each verb lemma yields 28 simple inflected forms, 14 simple 'written' inflected forms and 14 'spoken' inflected forms, and relatively, for each stem there are 7 inflected forms. The recognition of compound tenses is carried out by a set of 22 FST using the system morphological mode and the previous lexical annotation of simple verb forms. These FSTs allowed for the retrieval of 3,953 compound verb tenses from the corpus. For the evaluation of this language resource, a sample text, retrieved from the www.persian.euronews.com, and containing around 1000 words was used to assess the lexical coverage of the simple words' dictionary and the compound tenses' lexical graphs. The evaluation was done manually, based on the recognized words and those that were not identified by lexical resources built here.
    2014Master thesis Open access Show more