Maria-Mirela Petrea, Dan Cristea * Dealingwith Prosody. A Computer-Assisted Language Learning Approach

Fig. 12.- Anannotation example.

Tagging is done according to the following grammar:

<utterance>::= <list_of_adnotated_segms> |
<list_of_adnotated_segms> <silence_segm>
<list_of_adnotated_segms>::= <adnotated_segm> |
<adnotated_segm> <list_of_adnotated_segms>
<adnotated_segm>::= <text> (<list_of_adjacent_pitch_segms>) |
<silence_segm> <text> (<list_of_adjacent_pitch_segms>)
<list_of_adjacent_pitch_segms>::= <adjacent_pitch_segm> |
<adjacent_pitch_segm> <pitch_interuption> <list_of_adjacent_pitch_segms>
<adjacent_pitch_segms>::= <list_of_pitch_segms> <pitch_level>
<list_of_pitch_segms>::= <pitch_segm> |
<pitch_segm> <list_of_pitch_segms>
<pitch_segm>::= <pitch_level> <pitch_move> <loudness_move>
<pitch_level> ::=W| L| M| H| X| Z
<pitch_move> ::= S| R| /| F| \
<loudness_move> ::= *| #| !| ?| +| -
<pitch_interruption>::= <silence> | <unvoiced>
<silence> ::= =
<unvoiced> ::= _
<text> ::= any string of letters, but also ' and blank

Following is the annotation result of the phrase "I'd like cream" uttered by a male speaker:

=[23.23]I'd(_[9.62]MS*[93.78]M_[40.93]LS#[14.20]LR+[62.32])
like(LR+[80.25]MS-[104.43]L_[82.41])
=[155.61]cream(_[56.44]LR+[48.08]MF-[302.13]L_[14.06])

4. Conclusions and further work

The paper presented PROSODICS, an application dealing with the problem of computer-assisted language pronunciation learning. Its realization starts from the premise of the importance of prosody (intonation, stress and rhythm) in CALL. So far, PROSODICS was used for English teaching, but it could be easily extended for other languages. In fact its only link with a certain language is the phonemes features table, and perhaps also the font used for the graphical representation of the texts. Techniques that deal with prosody in artificial systems are yet scarcely described in the literature (see [1, 2, 11]). The paper shows how a bundle of elementary speech processing techniques can be used with spectacular results in a CALL application.

PROSODICS has two main functionalities:

• it helps in the creation of annotated speech data bases allowing a developer to either record or import utterances produced by native speakers of the target language, to attach the corresponding text to them in the orthographic or the phonemic transcription form, and to tag speech records on pitch/loudness criteria for laboratory studies;
• it assists a student to improve her pronunciation in the target language by displaying in an easily understandable format a comparison between the master's and her own utterance, where duration and rhythm, prosodic melody and stress are put in evidence. Exercises that concentrate on word level stress or sentence level stress can be imagined. At the word level, rules of words' accentuation can be devised (in a language as Italian, for instance, where most parts of the words receive the accent on the last but one syllable, the exceptions could be an easy trap for beginners; for a language rich in tonalities as Chinese the graphical facilities that PROSODICS offers for observing tones' movements could be of valuable help). At the sentence level, final intonation, as well as the melodic line of the utterance could be subjects of special exercises. Using PROSODICS, special training applications for actors or singers should be easily projected.

Acknowledgments. PROSODICS was implemented by a team formed by the authors together with Ciprian Bacalu, at the Department of Computational Linguistics of the University of Venice. We thank Rodolfo Delmonte, the head of the Laboratory of Computational Linguistics at the University of Venice for good critics, comments and suggestions. We are also indebted to Francesco Stiffoni for his permanent disposability to answer the developers questions regarding tricks of the Macintosh environment.

161