Accepted Papers


  • Automatic construction of multidocument Sunthesized sheets based on semantic categories in scientific papers
    1Olfa Makkaoui ,2Leila Makkaoui ,3Descles Jean-Pierre ,LaLIC-STIH ,France1,Institut du Cerveau et de la Moelle epiniere (ICM),France2,LALIC ,France3
    This paper presents a text mining tool for scientific publications that allows the extraction of textual segments (section, paragraph, sentences…) from a large corpora according to a set of semantic categories ( results, methods, hypothesis…). The extracted information is grouped according to their semantic affiliation which allows to obtain an organized textual representation called multi-document synthetized sheets. The automatic construction of these synthetized sheets is realized by semantically annotating documents according to a set of semantic categories. In fact, the annotation task is performed automatically using the Contextual Exploration processing (EC) that uses a set of linguistic markers associated with semantic categories.
  • Diagnosing learner errors on Arabic closed-class items
    Hayat Alrefaie and Allan Ramsay, The University of Manchester, United Kingdom
    Language learners are faced with a myriad of tasks: they have to learn how to make the sounds of the language they are learning, they have to learn its grammatical patterns, and they have to learn its vocabulary.There are numerous computational tools to help with the first two of these tasks. Providing help with the third raises a number of new challenges. The current paper describes a tool which will help learner to understand how to use ‘closed-class’ lexical items: this is a particularly taxing problem. To learn that the Arabic for ‘office’ is I.J ºÓ(mktb ) is fairly straightforward: keep saying to yourself ‘office’ = I.J ºÓ(mktb ) over and over again until it sticks. But learning the Arabic equivalent of ‘on’ is more difficult. The paper outlines a mechanism for providing diagnostic information about particular examples, with the aim of helping the learner to understand why a particular translation of a given closed-class item is appropriate in one situation but not in another.
  • Phonetic Features Of Malayalam Language For Speech Recognition Technology
    Cini Kurian, Cochin University of Science & Technology, Cochin
    Automatic speech recognition technology has achieved tremendous progress over the last 60 years. But performance comparable to that of humans cannot be achieved due to many factors. One of the reasons being the lack of study of phonetic realization of phonemes of the language, since speech recognition technology is highly dependent on the spoken form. Instead of depending highly on the written script, spoken form of each phoneme, especially unique phonemes in a language has to be analyzed well before designing a speech recognition system for a particular language.

    In this paper acoustic and phonetic analysis of one of the unique phonemes (Alveolar plosive) of Malayalam language along with its speech recognition accuracy have been presented. For speech recognition, the famous statistical classifier Hidden markov model is being used. The speech database has 32 selected minimal pairs spoken by 25 speakers. An analysis has been carried out to find out the resemblance of spectrogram, waveform, and format frequencies of the confusing pairs of the target word pairs. The objective of this paper is to find out whether the resemblance of these analysis will lead to misclassification in speech recognition and whether there is any relation between phonetic study and their speech recognition performance. Keywords: Speech recognition, unique phonemes, Malayalam.


Copyright ® AIAPP 2015