Taalmodel

'n Taalmodel is 'n waarskynlikheidsverdeling van ‘n volgorde van woorde.^[1] Gegewe enige volgorde van woorde van lengte $m$ , sal ‘n taalmodel ‘n waarskynlikheid $P(w_{1},\ldots ,w_{m})$ aan die hele volgorde toewys. Taalmodelle genereer waarskynlikhede deur opleiding op tekskorpusse in een of baie tale. Gegewe dat tale gebruik kan word om 'n oneindige verskeidenheid geldige sinne uit te druk (die eienskap van digitale oneindigheid), staan taalmodellering voor die probleem om nie-nul waarskynlikhede toe te ken aan linguisties geldige rye wat dalk nooit in die opleidingsdata teëgekom kan word nie. Verskeie modelleringsbenaderings is ontwerp om hierdie probleem te oorkom, soos die toepassing van die Markov-eienskap of die gebruik van neurale argitekture soos herhalende neurale netwerke of transformators.

Taalmodelle is nuttig vir 'n verskeidenheid probleme in rekenaarlinguistiek; vanaf aanvanklike toepassings in spraakherkenning^[2] om te verseker dat onsinnige (d.w.s. lae-waarskynlikheid) woordreekse nie voorspel word nie, vir wyer gebruik in masjienvertaling^[3] (bv. evaluasie van kandidaatvertalings), natuurlike taalgenerering (generering van meer mensagtige teks), woordsoortetikettering, ontleding,^[3] optiese karakterherkenning, handskrifherkenning,^[4] grammatika-induksie,^[5] inligtingherwinning,^[6]^[7] en ander toepassings.

Taalmodelle word gebruik in inligtingherwinning in die navraagwaarskynlikheidsmodel. Daar word 'n aparte taalmodel met elke dokument in 'n versameling geassosieer. Dokumente word gerangskik op grond van die waarskynlikheid van die navraag Q in die dokument se taalmodel: $M_{d}$ : $P(Q\mid M_{d})$ . Gewoonlik word die unigram-taalmodel vir hierdie doel gebruik.

Verwysings[wysig | wysig bron]

↑ Jurafsky, Dan; Martin, James H. (2021). "N-gram Language Models". Speech and Language Processing (3rd uitg.). Besoek op 24 Mei 2022.
↑ Kuhn, Roland, and Renato De Mori. "A cache-based natural language model for speech recognition." IEEE transactions on pattern analysis and machine intelligence 12.6 (1990): 570-583.
↑ ^3,0 ^3,1 Andreas, Jacob, Andreas Vlachos, and Stephen Clark. "Semantic parsing as machine translation." Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). 2013.
↑ Pham, Vu, et al. "Dropout improves recurrent neural networks for handwriting recognition." 2014 14th International Conference on Frontiers in Handwriting Recognition. IEEE, 2014.
↑ Htut, Phu Mon, Kyunghyun Cho, and Samuel R. Bowman. "Grammar induction with neural language models: An unusual replication." arXiv preprint arXiv:1808.10000 (2018).
↑ (1998) "A language modeling approach to information retrieval" in Proceedings of the 21st ACM SIGIR Conference.: 275–281, ACM. doi:10.1145/290941.291008.
↑ Hiemstra, Djoerd (1998). "A linguistically motivated probabilistically model of information retrieval" in Proceedings of the 2nd European conference on Research and Advanced Technology for Digital Libraries.: 569–584, LNCS, Springer. doi:10.1007/3-540-49653-X_34.

[1] Jurafsky, Dan; Martin, James H. (2021). "N-gram Language Models". Speech and Language Processing (3rd uitg.). Besoek op 24 Mei 2022.

[2] Kuhn, Roland, and Renato De Mori. "A cache-based natural language model for speech recognition." IEEE transactions on pattern analysis and machine intelligence 12.6 (1990): 570-583.

[Semantic_parsing_as_machine_translation-3] 3,0 ^3,1 Andreas, Jacob, Andreas Vlachos, and Stephen Clark. "Semantic parsing as machine translation." Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). 2013.

[4] Pham, Vu, et al. "Dropout improves recurrent neural networks for handwriting recognition." 2014 14th International Conference on Frontiers in Handwriting Recognition. IEEE, 2014.

[5] Htut, Phu Mon, Kyunghyun Cho, and Samuel R. Bowman. "Grammar induction with neural language models: An unusual replication." arXiv preprint arXiv:1808.10000 (2018).

[ponte1998-6] (1998) "A language modeling approach to information retrieval" in Proceedings of the 21st ACM SIGIR Conference.: 275–281, ACM. doi:10.1145/290941.291008.

[hiemstra1998-7] Hiemstra, Djoerd (1998). "A linguistically motivated probabilistically model of information retrieval" in Proceedings of the 2nd European conference on Research and Advanced Technology for Digital Libraries.: 569–584, LNCS, Springer. doi:10.1007/3-540-49653-X_34.

[1]

[2]

[3]

[4]

[5]

[6]

[7]