Text, Speech and Dialogue: 17th International Conference, TSD 2014, Brno, Czech Republic, September 8-12, 2014, Proceedings

... 7 7 7 7 9 9 9 9 10 11 ae: ae: n: n: n: n: n: n: n: n: n: n: n: t: t: t: t: t ... FST can be eliminated, or at least reduced, by first encoding it as a finite ... back as a FST. However, this redundancy-reduction procedure via the FSA ...

... Error-Driven Learning and Natural Language Processing: A Case Study in Part of Speech Tagging. Computational ... 59–66 (1999) Quantification of Segmentation and F0 Errors and Their Effect on 524 D. Spiliotopoulos, G. Petasis, and G ...

... forecast application. We chose 10 random sentences from a weather report. The weather report originated from one of the web sites included in the database collection. Thus, the style of the sentences was close to the speech corpus, but ...

... proper translation. Bilingual resources usually suggest more than one sense for a source word and it is not a ... thesaurus depend on the thesaurus scope. Numbers or dates identification can be appropriate for some types of clustering ...

... Language Technologies, pp. 1–8 (2001) 7. Onyenwe, I.E., Uchechukwu, C., Hepple, M.: Part-of-speech tagset and corpus development for Igbo, an African. In: Proceedings of LAW VIII-8th Linguistic Annotation, Workshop 2014 in conjuction ...

... TDT-1 corpus or from on-line RSS feeds. The TDT-1 corpus was chosen because of its clean, consistent and relatively error-free nature, and the large body of TDT-1 research available. We use the standard vector space model, where ...

... dominant - submissive 2 0 0 0 # 1 # 2 # 3 # 4 # 5 # 6 # 7 # 8 annotator ID # 9 # 10 # 11 # 12 # 14 # 1 # 2 # 3 # 4 ... agreement with her partner's proposal . This could be dominant because the agreement dominated the direction of ...

... definition of exposure is based on perplexity. We show below how a similar line of reasoning can be utilized for an exposure measure on discriminative models. Another related notion to the problem of memorization is membership infer ...

... interview already worked in our basic experiment with random samples. Manual transcriptions for all patients, which are in progress, will allow the targeted selection of patient speech in the interview and thus a more accurate ...

... (Super-Unit Length) be the number of units consecutively chosen from the database to generate a synthesised utterance ... system2: Table 1. Unit selection example for a Catalan utterance Text Un hivern pl`acid Transcription [uniBERmplasit] ...

... distinction is between informative summaries, which can be used instead of the source, and indicative summaries ... download 6 http://ufal.mff.cuni.cz/pdt/Morphology. SummEC: A Summarization Engine for Czech 1 Introduction.

... test) Danish English German Swedish Swiss German Norwegian Slovenian Slovak 1.04 1.69 0.80 2.56 0.51 2.31 4.15 0.54 1.14 0.78 0.00 0.55 0.27 0.53 0.87 0.00 2.36(A1-3) 2.62(A1-3) 2.40(A1-3) 1.52(A1-3) 1.06(A1-3) 4.43(A1-6) 4.86(A1-6) 1.72(A1 ...

... InfP, noun phrases NP, prepositionaladjective phrases PrepAdjP, prepositional-nominative phrases PrepNP, clauses SentP and reflexive marker sie ̧. Arguments can be parametrised. AdvP has no parameters, the only parameter of AdjP and NP ...

... Unit/word OOV Vocab.size Unit/word OOV Word 4110456 1 2.3% 420520 1 9.9% α = 0.1 606106 1.1 0.2% 183803 1.6 0.5% α = 0.01 170368 1.2 0.1% 53667 2.0 0.2% α =0.001 34521 1.6 0.1% 11562 2.9 0.3%. 6. Results ... vocabulary that gave the best WER, ...

... (InfP), nominal phrases (NP), prepositional-adjectival phrases (PrepAdjP), prepositional-nominal phrases (PrepNP) and clauses (SentP). A special slot sie hosts the reflexive marker. Some slots are parametrised. In particular, the only ...

... sentences. Information questions almost always are formed by full sentences. Repair initiations are different. Clarification is formed mostly by phrases. Reformulation and misunderstanding use full sentences in half of the cases and phrases ...

... 40k and 60k words (OOV: out of vocabulary word rate; WER: word error rate; MAP: mean average precision) OOV WER MAP – Text 0.4475 20k general LM 6.77% 24.2% 0.3013 20k topic LM 3.27% 20.3% 0.3412 20k interpolated LM 3.27% 19.6% 0.3393 ...

... pronunciation dictionaries : - the Onomastica pronunciation dictionary consisting of 233.919 names ; for names with ... success rate. - S5 pronunciation dictionary : 8.480 words . 6 Results The phonemisation errors were determined by ...

... (matka.Addr) ((Jirka.ELID.Act) pˇrij ́ıt.Pat (dom ̊u.Dir) (vcas-Temp) (10) Rodice zádali Jirku nechodit tam. Parents ... Milan (Has-met Jirku?) Milan George?) Met-Masc. (12) (Milan.Act.ELID) potkat (Jirka.Pat.ELID) In cases (i) through ...

... T.V. [Program package breeding-oriented and biometrics genetic methods ... program//Gene manipulation in plant breeding. NY, London, 1984. P ... selection and evaluation of breeding material]. Kiev, 1980. P.100-106. (in Russian). Stuber C.W. ...

... security Agency : Social Security Administration The company must register for the social security registration (which is referred to in Czech law as “the Register of Employers”) within 8 days of the date when the first employee starts ...

... Nfdump - netflow processing tools (2013), http://sourceforge.net/projects/nfdump/ 6. Hofstede, R., Drago, I., Sperotto, A., Sadre, R., Pras, A.: Measurement artifacts in netFlow data. In: Roughan, M., Chang, R. (eds.) PAM 2013. LNCS ...

... A B R1 R2 R3 R4 R5 R6 R7 R12 R3 R4 R7 R5 R6 A B (a) (b) (c) (d) or contour generation, can mislead fracture detection . To overcome this problem, we propose a novel approach based on relaxed digital straightline segment (RDSS). In the ...

... C , Brain Map ( b ) 38 ° C , Brain Map ( c ) 50 ° C , Brain Map ( d ) 28 ° C , Cross- Correlation Map ( e ) 38 ° C , Cross- Correlation Map Fig . 7. Correlation among Different Channels in the Field of Theta Wave according to the ...

... GuitarPro 7. Participants were instructed to focus on the quality of the music composition and not on the quality of the virtual instruments or the music production mix when answering questions. Participants. We recruited 31 ...