|
| << vorige | volgende >> |
Tijdschrift beschrijving
Alle jaargangen van het bijbehorende tijdschrift
Alle afleveringen van het bijbehorende jaargang
Alle artikelen van de bijbehorende aflevering
Details van artikel 2 van 5 gevonden artikelen
| Titel: |
DATA MINING FOR CREATING ACCENTUATION RULES |
| Auteur: |
Tomaz, Sef |
| Verschenen in: |
Applied artificial intelligence |
| Paginering: | Jaargang 18 (2004) nr. 5 pagina's 395-410 |
| Jaar: |
2004-01 |
| Inhoud: |
One of the important tasks in language analysis, speech synthesis, and recognition is stress assignment within words. The task is particularly difficult in languages where lexical stress can be located almost arbitrarily on every syllable in the word, such as in the Slovenian language. Although humans may pronounce very well even words they have never heard before, human experts have not been able to synthesize accurate rules for accentuation in phonetically complex languages. We have performed several experiments with different machine-learning and data-mining tools to create accentuation rules for the Slovenian language. We wanted to find out if it is possible to design better accentuation rules than the ones defined by the experts, and to test if these rules are complex or compact. The MULTEXT-East Slovene Lexicon was used for data sets, supplemented with lexical stress marks. The accuracy achieved by decision trees significantly surpassed all previous results. However, the sizes of the trees indicate that accentuation in Slovene is a very complex problem. We were not able to find a simple solution in the form of relatively compact rules using several ML and DM systems. It also indicates that creating accurate accentuation rules is a task that might require artificial intelligence (AI). |
| Uitgever: |
Taylor & Francis |
| Bronbestand: |
Elektronische Wetenschappelijke Tijdschriften |
Details van artikel 2 van 5 gevonden artikelen
| << vorige | volgende >> |