ID Author name E-mail Authors Title Topic Abstract file Rev. 1 Rev. 1 ranking Rev. 1 comment Rev. 2 Rev. 2 ranking Rev. 2 comment Accepted as PDF file PS file Corrected paper Paper finished Comment
1 Hisao Kuwabara 1 Perceptual Properties of Syllables Isolated from Continuous Speech for Different Speaking Rate Speech - speech segmentation 1_Hisao_Kuwabara_abstract.txt Hynek Hermansky    Genevieve Baudoin    Reject 1_Hisao_Kuwabara.pdf         
2 Shu-Chuan Tseng 1 Transcribing and Annotating Mandarin Conversational Dialogues Speech - other 2_Shu-Chuan_Tseng_abstract.txt Attila Ferencz    Nikola Pavesic    Reject 2_Shu-Chuan_Tseng.pdf         
3 Jorge Grana and Miguel A. Alonso and Manuel Vilares 3 A Common Solution for Tokenization and Part-of-Speech Tagging Text - parsing and part-of-speech tagging 3_Jorge_Grańa_abstract.txt Karel Oliva    Eva Hajicova  An interesting (and useful) topic, with an interesting solution. It would be interesting to see how the solution applies not only to ambiguous structures as those quote in the paper but in a more general environment  LP   3_Jorge_Grań       
4 Vlastislav Dohnal 3 Approximate Searching in Large Collections of Text Data Text - information retrieval 4_Vlastislav_Dohnal_abstract.txt Hanks Patrick    Jaroslava Hlavacova  I would prefer to include the definition of the D-index, which appears to be the key term of the paper, to the definition of such a common thing as metric space. Without it I couldn't understand the section 3 of the article. Also the basic school artithmetic ("... a sequential search on 50,000 sentences takes about 5 minutes, which would result in a half an hour for a 6 times larger file of 300,000 sentences.") is not necessary to put into article. The definition of "edit transformation" seems to be not complete - I miss the relationship between the strings X, Y and the sequence S (probably X=S1, Y=Sn ?) What is "distance density", "read" as a noun?   Reject 4_Vlastislav_Dohnal.pdf       
5 Ilya Oparin 3 Differences in the Stability of Russian and Greek Speech Signal towards Noise Effects Speech - other 5_Ilya_Oparin_abstract.txt Josef Psutka  Interesting theme. But performed experiments are described very slightly and it is not evident if obtained results can be generalized. Experiments with Greek were executed under different conditions and can be only roughly compared with those done for Russian.   Leon Rothkrantz  -The authors want to report about 4 problems in one paper, the work seems very preliminary -the use of statistics is not sufficient, the experimental design is very poor -the researchers choose sentences, instead of isolated words, but then there is a great mpact of context -Greek people living most of the time in petersburg or in Patras are in more then one aspect very different, it is not clear why the greek language is considered and why the reserachers don't restrick themselves to the Russian language -the paper has no literature survey and no results from others are reported   Reject 5_Ilya_Oparin.pdf         
6 Amparo Varona 2 Integrating high and low smoothed LMs in a CSR system Speech - automatic speech recognition 6_Amparo_Varona_abstract.txt E.G. Schukat-Talamazzini    Pavel Skrelin    Reject 6_Amparo_Varona.pdf       
7 Imad A. Al-Sughaiyer and Ibrahim A. Al-Kharashi 2 Rule Parser for Arabic Stemmer Text - automatic morphology 7_Imad_A._Alsughaiyer_abstract.txt Eduard Hovy      Steven Krauwer    LP 7_Imad_A._Alsughaiyer.pdf         
8 Kristîne Levâne 1 Latvian Corpus Text - text/topic summarization 8_Kristîne_Levâne_abstract.txt Jaroslava Hlavacova  In the Introduction there are mentioned 30 million words, in the Conclusion 3 million.   Karel Oliva  The paper is called "Latvian Corpus" but its contents is mainly concerned= with Latvian morphology. So maybe you should consider an appropriate cha= nge of the title. The morphological analyzer of Latvian is, in addition, obviously only in = a very initial state. Accordingly, the paper presents interesting problem= s rather than solutions to them. =20 The section on XML and statistics report on a "research started only mont= h ago" (quoted form the paper) or "some attempts". I am afraid this is sl= ightly less than I would expect for TSD conference. In addition, this is in contrast with the statement that computer-aided i= nvestigation of Latvian started already in 1990. So I guess there must be= a lot of work completed, but this is somehow not mirrorred in the paper,= unfortunately. It would be also better if the Latvian examples had more complete English= translations. As for English, correct:  Reject   8_Kristîne_Levâ       
9 Tomaž Šef and Maja Škrjanc and Matjaž Gams 3 Automatic Lexical Stress Assignment of Unknown Words for Highly Inflected Slovenian Language Speech - text-to-speech synthesis 9_Tomaz_Sef_abstract.txt Taras Vintsiuk      Hynek Hermansky    LP 9_Tomaz_Sef.pdf       
10 Jesús Vilares 5 Practical NLP-Based Text Indexing Text - information retrieval 10_Jesús_Vilares_abstract.txt Vladimir Petkevic  The complex task you tackle needs other methods. For instance: HMM tagger is totally inadequate for Spanish (as for many other languages!). You should develop a rule-based tagger reflecting the system of Spanish and you need a big corpus. Without these two prerequisites you can hardly achieve good results. No substantial results are presented. You can hardly claim that your architecture is generally good enough so as to be applicable to other languages - this sounds very courageously (see the last sentence in the 1st par., page 7). You'd improve your English! There seem to be no new ideas with respect to existing methods in NLP.   Yorick Wilks      Reject   10_Jesú       
11 András Kocsor and Kornél Kovács 2 Kernel Springy Discriminant Analysis and its Application to a Phonological Awareness Teaching System Speech - other 11_Andras_Kocsor_abstract.txt Genevieve Baudoin    Attila Ferencz    SP 11_Andras_Kocsor.pdf         
12 Lukasz Debowski 3 Testing the Limits - Adding a New Language to an MT System Text - machine translation 12_Lukasz_Debowski_abstract.txt Dr. Alexander Gelboukh Kahn  Paper: Testing the Limits by Lukasz Debowski et al. Remove page number from 1st page.   Eneko Agirre    Reject 12_Lukasz_Debowski.pdf         
13 Kornel Kovacs 3 Hungarian Speech Synthesis Using a Phase Exact HNM Approach Speech - speech modeling 13_Kornel_Kovacs_abstract.txt Nikola Pavesic    Josef Psutka  Nice paper with new ideas. Maybe it could be extended and presented as oral.   Reject 13_Kornel_Kovacs.pdf         
14 Pavel Květoň and Karel Oliva 2 Achieving an Almost Correct PoS-Tagged Corpus Text - parsing and part-of-speech tagging 14_Pavel_Kveton_abstract.txt Eva Hajicova  The core idea of this paper is interesting and certainly valuable. However, there are two places in the paper the authors should reconsider: (a) pages 1 through 2 incl. are rather 'talkative' (Sections 0 through 3) and in some places repeat trivial considerations); they may be shortende, made more factual and thus save place for a more explicit (and exemplified) exposition in the following sections. (b) in Section 4the results should be formulated more carefully and explicitly, since it is only the collocation "within the test sections" that indicates that the authors are not so naive as to use the same data for training and testing! the present wording would make the whole experiment dubious.   Hanks Patrick    LP       
15 Rei Oguro and Hiromi Sekiya and Yuhei Morooka and Kazuyuki Takagi and Kazuhiko Ozeki 5 Evaluation of a Japanese Sentence Compression Method Based on Phrase Significance and Inter-Phrase Dependency Text - text/topic summarization 15_Rei_Oguro_abstract.txt Jaroslava Hlavacova    Eduard Hovy      LP 15_Rei_Oguro.pdf       
16 Tiit Hennoste 6 Determining Dialogue Acts in Estonian Dialogue Corpus Dialogue - other 16_Tiit_Hennoste_abstract.txt Ivan Kopecek    Vaclav Matousek  The work isn't original, but the topic seems me to be very useful for the developing former Soviet country. I recommend to give an opportunity to the authors to present this submission at the conference.  Reject 16_Tiit_Hennoste.pdf         
17 Michael V. Boldasov and Elena G. Sokolova and Michael G. Malkovsky 3 User query understanding by the InBASE system as a source for a multilingual NLG module(first step) Text - multi-lingual issues 17_Michael_V._Boldasov_abstract.txt Steven Krauwer    Elmar Noeth  interesting work Reconsider the title: the term NLG is not known in the communitiy and not introduced in the paper Could you elaborate a little on users' experiences? The English needs work! (Maybe you have a competent proofreader) stuff should be staff! also anybody, everybody,   SP       
18 Anton Batliner and Viktor Zeissler and Elmar Nöth and Heinrich Niemann 4 Prosodic Classification of Offtalk: First Experiments Dialogue - prosody and emotions in dialogues 18_Anton_Batliner_abstract.txt Ivan Kopecek    Karel Pala  ---  LP 18_Anton_Batliner.pdf       
20 Tomaž Rotovnik and Mirjam Sepesy Maučec and Bogomir Horvat and Zdravko Kačič, 4 Large Vocabulary Speech Recognition of Slovenian Language Using Data-Driven Morphological Models Speech - automatic speech recognition 20_Rotovnik_Tomaz_abstract.txt Leon Rothkrantz  -It is a pity that the authors did not submit an extended paper, providing more technical details -In ASP recognition of the "stem"of words is important, ďnflection"can be generated from the context using grammatical rules -Using syllables as modelling units seems to be a natural extension of the proposed method in the tradition of the French schools. But then many subwords models have to be trained and a huge number of data is necessary. This aspect should be discussed in more details -it is not clear why the authors used a reduced phoneme set -the English should be improved , many articles (the, a etc.) are deleted examples section 1 line 1: The Slovenian language is a highly ... line 4 have a complex morphological structure etc. -there is an error is tabele 1, 1665 should be replaced by 165 -in the abstract the authors claim that they got an improvement of 2.5 %, but in the conclusion thewy report about similar results?   E.G. Schukat-Talamazzini    SP 20_Rotovnik_Tomaz.pdf         
21 Jinyoung Kim 2 Modified LBG Clustering Algorithms for Small Unit Inventory in Corpus-based TTS system Speech - text-to-speech synthesis 21_Jinyoung_Kim_abstract.txt Pavel Skrelin    Taras Vintsiuk    Reject 21_Jinyoung_Kim.pdf         
22 Manolis Maragoudakis and Aristomenis Thanopoulos and Nikos Fakotakis 3 Statistical Decision Making applied to Text and Dialogue Corpora for Effective Plan Recognition Dialogue - development of dialogue strategies 22_Manolis_Maragoudakis_abstract.txt Ivan Kopecek    Vaclav Matousek  accept after removing minor errors in English  LP 22_Manolis_Maragoudakis.pdf         
23 Dana Hlavackova 2 Morphological Tagging of the Brno Spoken Corpus Text - text corpora 23_Dana_Hlavackova_abstract.txt Karel Oliva  It seems to me that there occurs quite a strong tension between the title= of the paper (Morphological tagging ....) and its contents: there is a l= ot of information about the corpus of spoken Czech, but virtually *nothin= g* about how the tagging is performed (and there is no hint even in the R= eferences). The *only* information a reader gets is that your tagging is somewhere be= tween 60% and 70% correct (once you say 60, once 70). But there is nothin= g about the method, about how the results were measured ... Also, I missed any information about your contribution: is it you who col= lected the data ? Is it you who annotated parts of them / invented and te= sted some new tagging method ... You should either change the contents of the paper accordingly (and profo= undly) or change the title, hence.   Vladimir Petkevic  The article is very short and not interesting, its is poor in content. It only sketchily describes the whole subject with no interesting ideas presented. The reported results are poor (also due to the complexity of the problem). The approach should have been described in a more detailed way, it should have concentrated on interesting aspects and report the results more thoroughly with the discussion of problems encountered.   Reject       
24 Marta Gatius and Horacio Rodríguez 2 NATURAL LANGUAGE GUIDED DIALOGUES FOR ACCESSING THE WEB Dialogue - dialogue systems 24_Marta_Gatius_abstract.txt Elmar Noeth  Fig. 1 is unreadable (prob. because it is a color picture) I'm missing information about user experiences   Karel Pala    LP 24_Marta_Gatius.pdf         
25 Jiří Mírovský 2 NetGraph System--Searching through the Prague Dependency Treebank Text - text corpora 25_Jiří_Mírovský_abstract.txt Yorick Wilks      Dr. Alexander Gelboukh Kahn  Paper: NetGraph System by Juri Mirovsky et al. Remove running page heads and page numbers.   Reject 25_Jiří_Mírovský.pdf         
26 Jindřich Matoušek and Daniel Tihelka and Josef Psutka and Jana Hesová 4 German and Czech Speech Synthesis Using HMM-Based Speech Segment Database Speech - text-to-speech synthesis 26_Jindrich_Matousek_abstract.txt Hynek Hermansky    Genevieve Baudoin    LP 26_Jindrich_Matousek.pdf         
27 Markéta Lopatková and Veronika Řezníčková and Zdeněk Žabokrtský, 3 Valency Lexicon for Czech: from Verbs to Nouns Text - other 27_Zdenek_Zabokrtsky_abstract.txt Eneko Agirre  Interesting paper, although preliminary. You might be interested in work on the dissambiguation of derivational suffixes and the meaning of the roots (check Penthedourakis, J. and Vanderwende, L., 1993. Automatically Identifying Morphological Relations in Machine-Readable Dictionaries. Microsoft internal report MSR-TR-93-06.).   Yorick Wilks  I cannot judge the Czech's but this seems an excellent and well-planned and motivated resource.  SP       
29 Georg Stemmer and Stefan Steidl and Elmar Nöth and Heinrich Niemann and Anton Batliner 5 Comparison and Combination of Confidence Measures Speech - automatic speech recognition 29_Georg_Stemmer_abstract.txt Attila Ferencz    Nikola Pavesic    LP 29_Georg_Stemmer.pdf       
30 Andrej Žgank and Tomaž Rotovnik and Zdravko Kačič and Bogomir Horvat 4 Uniform Speech Recognition Platform for Evaluation of New Algorithms Speech - automatic speech recognition 30_Andrej_Zgank_abstract.txt Josef Psutka  The uniform platform is a good thing to provide many comparative experiments in spoken Slovenian. The performed tests could be considered as the standard work.   Leon Rothkrantz  -the research project should be reported as an extended paper, no a lotof technical details are missing -the architecture could be presented using a Figure with the essential components -it is not clear what is implemented and what and how it can be tested, a lot is reported in general terms -what is an evaluation of implementation"" ??, how evaluation is implemented?? evaluation of the implemented system?? the goal of this section is to show thatb the tool can be used and the results of a testexample are reported? In that case more details should be presented how to use the system -the text has a lot of spelling errors (articles ""ä", "the") examples abstract line 1 the development of a speech recognition platform introduction line 5 is based on the similar ... line 7 for a broad spectrum .... line 10 in the future the port to the ...   SP 30_Andrej_Zgank.pdf         
31 Jan Nouza 1 Strategies for Developing a Real-Time Continuous Speech Recognition System for Czech Language Speech - automatic speech recognition 31_Nouza_abstract.txt E.G. Schukat-Talamazzini    Pavel Skrelin    LP 31_Nouza.pdf         
32 Jan Nouza and Petr Kolář and Josef Chaloupka 3 Voice Chat with a Virtual Character: The Good Soldier Svejk Case Project Dialogue - dialogue systems 32_Jan_Nouza_abstract.txt Ivan Kopecek    Vaclav Matousek    SP 32_Jan_Nouza.pdf         
33 I. Azzini and T. Giorgino and D. Falavigna and R. Gretter 4 Application of Spoken Dialogue Technology in a Medical Domain Dialogue - dialogue systems 33_Daniele_Falavigna_abstract.txt Elmar Noeth  Nice work; it would be great to see preliminary results of the field test in the final version of the paper   Karel Pala  The meaning of the abbreviation 'ITC-irst' referring to your Institute as far as I can see is not very obvious at the first glance - perhaps it would help to change the title of the paper a little? kp.   SP       
34 Goran Nenadić and Irena Spasić and Sophia Ananiadou 3 Term Clustering using a Corpus-Based Similarity Measure Text - knowledge representation and reasoning 34_Goran_Nenadic_abstract.txt Hanks Patrick    Jaroslava Hlavacova  Isn't there a (typing) mistake in the fraction denominators in the definitions of CS and LS? Content-Description: 34.pdf   SP 34_Goran_Nenadic.pdf         
35 Emilio Sanchis and Fernando García and Isabel Galiano and Encarna Segarra 4 Applying dialogue constraints to the understanding process in a Dialogue system Dialogue - dialogue systems 35_SANCHIS,_EMILIO_abstract.txt Ivan Kopecek    Vaclav Matousek  interesting for presentation, useful for the Spanish country; I recommend to accept this submission and to give to the authors to present this submission at the conference.  LP 35_SANCHIS,_EMILIO.pdf  35_SANCHIS,       
36 Carlos D. Martínez-Hinarejos and Francisco Casacuberta 2 Evaluating a Probabilistic Dialogue Model for a Railway Information Task Dialogue - other 36_Carlos_D._Martinez-Hinarejos_abstract.txt Elmar Noeth  Nice work; of course not enough data :) Personally I would cite Searle rather than Fukuda et al. when introducing dialogue acts. There has also been earlier and more profound work on dialogue acts and speech recognition. If I interpret your numbers on page 5 correctly, then in 194 dialogs only 174 system turns were uttered? This does not make sense! Please clarify what number is Type and what is Token Look at: M. Boros, W. Eckert, F. Gallwitz, G. Görz, G. Hanrieder, and H. Niemann. Towards understanding spontaneous speech: Word accuracy vs. Concept accuracy. In Proc. Int. Conf. on Spoken Language Processing, volume 2, pages 1005-1008, Philadelphia, PA, USA, 1996. where - rather than on the speech act level - a system accuracy of a train information system is calculated on the level of semantic entities. May be that is more appropriate. (The paper per se deals with recognition errors, but the definition of concept accuracy might be interesting for you)   Ivan Kopecek    LP 36_Carlos_D._Martinez-Hinarejos.pdf       
37 Dana Nejedlová, 1 Comparative Study on Bigram Language Models for Spoken Czech Recognition Speech - automatic speech recognition 37_Dana_Nejedlova_abstract.txt Taras Vintsiuk    Hynek Hermansky    LP 37_Dana_Nejedlova.pdf         
38 Pascal Wiggers and Leon J. M. Rothkrantz 2 Integration of speech recognition and automatic lipreading Speech - automatic speech recognition 38_Pascal_Wiggers_abstract.txt Genevieve Baudoin    Attila Ferencz    LP       
39 Dimitri Woei-A-Jin 2 Anaphora Resolution in a speech recognition environment Speech - other 39_Dimitri_Woei-A-Jin_abstract.txt Nikola Pavesic    Josef Psutka  Very nice application.   will not arrive       
41 Vlasta Radova 2 A Method for Segmentation of Voiced Speech Signals into Pitch Period Segments Speech - speech segmentation 41_Vlasta_Radova_abstract.txt Leon Rothkrantz  At this moment there are many methodsproposed for automatic segmentation, but none of them is perfect. In the paper an iteresting new approach is presented. But the paper has to minor points First the authors don't relate their work to the work of others, they even don't mention similar approaches by other (only Vintsiuk) Seconly it would be nice to compare the test results with wellknown segmentationalgorithm, so to implment both and test it on the same corpus Thirdly I don't see why the testresuts are not compared with mauasegmentation. Take for example the TIMIT database which is segmented and apply the new segmentationalgoritm on that database (I presume that the proposed algorthm is language independent on the voiced part of the speech Automatic selection of voiced part ofspechca be realised by autocorrelation methods.   E.G. Schukat-Talamazzini    Reject 41_Vlasta_Radova.pdf       
42 Michal Prcín and Luděk Müller 2 Heuristic and Statistical Methods for Speech/Non-speech Detector Design Speech - automatic speech recognition 42_Michal_Prcin_abstract.txt Pavel Skrelin    Taras Vintsiuk    LP       
43 Adam Przepiorkowski 2 The Unbearable Lightness of Tagging: Case Study in Polish Morphology Text - parsing and part-of-speech tagging 43_Adam_Przepiorkowski_abstract.txt Eduard Hovy      Steven Krauwer  What should come out of a tagger is solely dependent on the needs of the process that is going to use the result. If that process is not known or not properly defined, the whole discussion becomes pretty empty.   Reject 43_Adam_Przepiorkowski.pdf         
44 Nestor Garay-Vitoria and Julio Abascal and Luis Gardeazabal 2 Evaluation of prediction methods applied to an inflected language Dialogue - assistive technologies based on speech and dialogue 44_Nestor_Garay-Vitoria_abstract.txt Ivan Kopecek    Vaclav Matousek  interesting submission; I reccomend to accept it as is. secondly: I reccomed to submit this article to the conference on minor languages too.  LP       
45 Andrés Montoyo and Rafael Romero and Sonia Vázquez and Carmen Calle and Susana Soler 5 The Role of WSD for Multilingual Natural Language Applications Text - word sense disambiguation 45_Andrés_Montoyo_abstract.txt Steven Krauwer    Karel Oliva  The paper proposes a new architecture, but contains only very little comparison to other systems approaches. Also, it completely lacks any evaluation (how successful the system is. There are quite some erors in English, in particular in subject-verb agreement (and also some others).   LP   45_André       
46 Maria Yavorskaya, 3 Wordnet as a Tool for Measurement of Domain Similarity of Texts Text - information retrieval 46_Maria_Yavorskaya_abstract.txt Vladimir Petkevic  The article is but a sketch. It is written in a very negligible and general way. No results have been presented, nothing interesting shown.   Yorick Wilks      Reject 46_Maria_Yavorskaya.pdf         
47 Antoine Rozenknop 1 A Gibbsian Context-Free Grammar for Parsing Text - parsing and part-of-speech tagging 47_Antoine_Rozenknop_abstract.txt Dr. Alexander Gelboukh Kahn  Paper: A Gibbsian Context-Free... by Antonie Rozenknop My low evaluation of the technical quality of your paper is due to insufficient experimental results: you did not demonstrate that your approach gives good results in the real life situation (Test not= Learn). I am sure your paper would be unconditionally accepted if you presented real evaluation and demonstrated real improvement on the unseen data. Also, your paper definitely needs proof-reading by someone with good knowledge of English. Some words look like written in French (and some even _are_ written in French). Title: consider removing "A": "Gibbsian..." Author: Where is Footnote 1? Address: Suisse = Switzerland? Please use Englihs! Consider the following changes: Abstract:   Karel Oliva  The paper puts forward an interesting discussion on an alternative to PCFG, and above all proposes such an alternative. Generally, the submission seems a little bit premature - it woudl be definitely nicer if you were able to give more evaluation as well as could report on the results of the work you say you only plan. Then the paper would be truly excellent. So I consider the contents of the paper very good - what is the problem is the language. If accepted you DEFINITELY have to turn the Franglais into real English !!! On very many places, you use words which are kind of mixed, sometimes you even use pure French (the article "la", should be "the", or captions to Fig 2).   LP 47_Antoine_Rozenknop.pdf         
48 Ilyas Potamitis and Nikos Fakotakis and Nikos Liolios and George Kokkinakis 1 SPEECH ENHANCEMENT USING MIXTURES OF GAUSSIANS FOR SPEECH AND NOISE Speech - other 48_Potamitis_Ilyas_abstract.txt Hynek Hermansky    Genevieve Baudoin    SP 48_Potamitis_Ilyas.pdf         
49 Armando Suárez and Manuel Palomar 2 Word Sense vs. Word Domain Disambiguation: a Maximum Entropy approach Text - word sense disambiguation 49_ARMANDO_SUAREZ_abstract.txt Eva Hajicova    Hanks Patrick    LP 49_ARMANDO_SUAREZ.pdf         
50 César González Ferreras and David Escudero Mancebo and Valentírn Carde\ noso Payo 3 From HTML to VoiceXML: A first approach. Dialogue - markup languages related to speech and dialogue 50_Cesar_Gonzalez_Ferreras_abstract.txt Elmar Noeth  nice paper Do you have any user experiences? Please elaborate on chapter 3: how do you construct the FSD. Having in mind that Vxml and your system will be used, can this influence the original design of a web-page? How much hand work   Karel Pala  -- kp.   SP       
51 Juan Luis Garcia Zapata 4 On Portability of Automatic Speech Recognition: A Study Case Speech - automatic speech recognition 51_Juan_Luis_Garcia_Zapata_abstract.txt Attila Ferencz    Nikola Pavesic    Reject       
52 Piotr Banski 1 The Pros and Cons of Stand-off Annotation: IPI PAN Corpus Design Text - text corpora 52_Piotr_Banski_abstract.txt Jaroslava Hlavacova    Eduard Hovy      Reject 52_Piotr_Banski.pdf         
53 D.W. Oard and D. Demner-Fushman and J. Hajič and B. Ramabhadran and S. Gustman and W.J. Byrne and D. Soergel and B. Dorr and P. Resnik and M. Picheny 10 Cross-Language Access to Recorded Speech in the MALACH project Text - information retrieval 53_Douglas_W._Oard_abstract.txt Steven Krauwer    Leon Rothkrantz    LP 53_Douglas_W._Oard.pdf       
54 Gies Bouwman and Lou Boves 2 Utterance Verification based on the Likelihood Distance to Alternative Paths Speech - automatic speech recognition 54_Gies_Bouwman_abstract.txt Josef Psutka  Very nice paper with many original ideas.   Leon Rothkrantz  -the authors report about improvements in recognition rate, they also report about two classes of error substitution and insertion for which class they got the best results? -it proves that Cart is better than LC, as can be expected, but no evidence is reported -it is not clear if the corpus consists of isolated words of city names , maybe those names are extracted from sentences of continuous speech recordings -there is a typing error in the introduction line 14 on the based on speech??   LP 54_Gies_Bouwman.pdf         
55 Tomáš Bartoš and Luděk Müller 2 Rejection technique based on the mumble model Speech - other 55_Tomáš_Bartoš_abstract.txt E.G. Schukat-Talamazzini    Pavel Skrelin    LP 55_Tomáš_Bartoš.pdf         
56 Petr Motlírček and Lukáš Burget 2 Efficient Noise Estimation and its Application for Robust Speech Recognition Speech - automatic speech recognition 56_Petr_Motlicek_abstract.txt Taras Vintsiuk      Hynek Hermansky    LP 56_Petr_Motlicek.pdf       
58 Milan Sečujski and Radovan Obradović and Darko Pekar and Ljubomir Jovanov and Vlado Delić 1 Synthesis in Serbian Language Speech - text-to-speech synthesis 58_Milan_Secujski_abstract.txt Genevieve Baudoin    Attila Ferencz    LP 58_Milan_Secujski.pdf         
59 Gregers Koch 1 Basic Principles of Automated Information Extraction Text - information retrieval 59_Gregers_Koch_abstract.txt Karel Oliva  To call the paper "BASIC PRINCIPLES of Automated Information Extraction" = seems to be a kind of overkill. =20 The paper in fact proposes to look at (extracting) informational content = as onto a dataflow (Sect. 3). =20 And it proposes the implementation of this dataflow by means of sharing t= he logical variable. =20 I am afraid that such a view is neither particularly original nor complex= enough to take care of the problems which arise in information extractio= n from real texts. In fact, exactly the same concept (aka "semantic repre= sentation") has been proposed already in the first papers presenting Defi= nite Clause Grammars about 20 years ago. Apart from this general view, some minor points: - whether assigning one or several formalized semantic representations is= "An absolutely central problem of semantics" is a matter of personal vie= w. Maybe a slightly less radical formulation might be due.  Vladimir Petkevic  The extraction of information content is an extremely complex task and I think your approach, as far I could understand it, can hardly work for only slightly more complex sentences than the one you presented. The overall approach is, to my mind, very naive and can hardly have significance which surpasses the realm of only the simplest sentence structures. I do not claim, however, to be an expert in the area. Your English should be better - for instance, the 1st sentence in the abstract is syntactically wrong. Your presentation, esp. the trees are almost illegible.   Reject       
60 Krasimira Petrova 2 Adaptation of Swedish Transcription System for Spoken Language Analysis for Bulgarian Speech - other 60_Krasimira_Petrova_abstract.txt Nikola Pavesic    Josef Psutka  This is not a scientific paper but only a short report on a bilateral project "Multimedia ... " without a scientific background.   Reject 60_Krasimira_Petrova.pdf         
61 Marek Trabalka and Mária Bieliková 2 Using Salient Words to Perform Categorization of Web Sites Text - information retrieval 61_Marek_Trabalka_abstract.txt Yorick Wilks      Dr. Alexander Gelboukh Kahn  Paper: Using Salient Words.. by Marek Trabalka et al. Your paper definitely needs proof-reading by someone with good knowledge of English. The use of articles in your text is a disaster! Sometimes it is just unreadable because of wrong use of articles.   LP 61_Marek_Trabalka.pdf         
62 Petya N. Osenova 2 automatic Identification and Linguistic Description of the Abbreviation in the BulTreeBank Electronic Text Archive Text - parsing and part-of-speech tagging 62_Petya_N._Osenova_abstract.txt Eneko Agirre  The approach is interesting, but basic. There is no comparison to other work on the area (Park & Byrd, (Sundaresan & Yi,, which could be used to improve the system. Section 2 contains too many references. In section 4.2 the statistical criterion is difficult to understand. Better show an example here. Specifically TokPar is not clear.   Eva Hajicova    Reject       
63 Pavel Skrelin 2 A Physical Pause as a Sequence of Special Articulation Gestures Speech - other 63_Pavel_Skrelin_abstract.txt Leon Rothkrantz  The authors claim that the results are different from other studies , but those differences can be caused by other reasons such as differences in testmaterial, subjects etc. the results in the tables are not convincing. It is difficult to use statistical test with such a small number of testpersons but then it is clear if the difference is significant or not in an objective way. If we take recordingd of a newsreader the the same results can be expected? -from table 1 and others we conclude that there is a lot of variation between subjects, there is no explanation for that. -it is not clear why pauses are labeled as "psychological -in section 3 the authors report that a great majority ... (how great is great??) -there are many spelling errors in the text (the articles "the: and ä ëxample section 1 line 1 speech recognition system has an algorithm line 5 we know that speakers line 14 boundaries in a Russian text -the paper has no abstract, section 2 is losely connected to the rest of the paper   E.G. Schukat-Talamazzini    Reject 63_Pavel_Skrelin.pdf         
64 Darko Pekar pekard@EUnet.yu 3 ALFANUM SYSTEM FOR CONTINUOUS SPEECH RECOGNITION Speech - automatic speech recognition 64_Darko_Pekar_abstract.txt Pavel Skrelin    Taras Vintsiuk    Reject 64_Darko_Pekar.pdf         
66 Roman V. Mescheriakov 3 RECOGNITION AND SPEECH SYNTHESIS IN THE DIALOGUE SYSTEMS STRUCTURES Dialogue - other 66_Roman_V._Mescheriakov_abstract.txt Ivan Kopecek    Vaclav Matousek  I didn't see this approach; from this viewpoint the article seems me to be original; but the presentation contains a lot of general phrases; is this work really original? For all that I recommend to give to the author an oportunity to present the submission on the conference.  Reject 66_Roman_V._Mescheriakov.pdf         
67 Bronius Tamulynas 1 Multilingual Computer-based Communication and Language Processing: Lithuania case Text - multi-lingual issues 67_Bronius_Tamulynas_abstract.txt Hanks Patrick    Jaroslava Hlavacova  If the aim was to describe a general strategy of computer based translation from one language into another, it is too brief, naming just old common points without going into details. If you wanted to tell something about your special experience with one couple of languages - English / Lithuanian, I would expect a factual data.   Reject 67_Bronius_Tamulynas.pdf         
68 Nira B. Volskaya 1 Pause duration at syntactic boundaries Speech - other 68_Nira_B._Volskaya_abstract.txt Hynek Hermansky    Genevieve Baudoin    will not arrive 68_Nira_B._Volskaya.pdf         
69 Huang Ke, Ma Shaoping 2 Text Categorization Based On Concept Indexing and Principal Component Analysis Text - other 69_Huang_Ke,_Ma_Shaoping__abstract.txt Eduard Hovy      Steven Krauwer    will not arrive 69_Huang_Ke,_Ma_Shaoping_.pdf         
70 Gábor Alberti and Helga M. Szabó 2 Discourse-Semantic Analysis of Hungarian Sign Language Text - lexical semantics and semantic networks 70_GÁBOR_ALBERTI__abstract.txt Eva Hajicova  A very nice paper, focussing on an issue that is not that often discussed at this type of conference but that deserves full attention. It would be also interesting to see how the information-structure analysis (topic-focus articulation) would enrich the DRS's for this application.   Karel Oliva  First, the paper does not conform to the required format (e.g., there is no Abstract). Further, it seems to be kind of "torn out" of a larger paper, in fact being something like the first half of a paper. It has no evaluation and worse even no comparison to other methods, no conclusion section ... You should make the paper look like a stand-alone, complete work.   LP 70_GÁBOR_ALBERTI_.pdf         
71 Janez Žibert and France Mihelič and Nikola Pavešić 3 Speech Features Extraction Using Cone-shaped Kernel Distribution Speech - automatic speech recognition 71_Janez_Zibert_abstract.txt Attila Ferencz    Leon Rothkrantz  -The topic of the paper is very interesting but it is in the borderline of TSD, but fits better in ICASS or related signal processing cnferences -In fig 1a and 1b authors compare the results of spectrogram analysis and CKD. But as is noticed the spectrogramresults could be better if the gal is to analyse frequencies (instad of better time resolution. In fig b CKD is optimised for frequency analysis, so the results are very good. In the last section it proves that CKD is etter i recognition of vowels (voiced part of speech) instead of SPEC. We expect that spectogramanalysis would be better on the voiced part of the speech. We also hoped that CKD was better on the nonvoiced part of the speech because common techniques are far from optimal in that area. The kernel fuction approach is related to the "wavelet"-approach but that is not mentioned in the paper   LP 71_Janez_Zibert.pdf         
72 Simon Dobrišek and Jerneja Gros and Boštjan Vesnicer and France Mihelič and Nikola Pavešić 5 A Voice-Driven Web Browser for Blind People Dialogue - dialogue systems 72_Simon_Dobrisek_abstract.txt Elmar Noeth    Karel Pala  The information given about the particular components of the described browser is rather general, it should be a bit more specific. kp.   LP 72_Simon_Dobrisek.pdf       
74 Josef Psutka and Pavel Ircing and Josef V. Psutka and Vlasta Radová and William J. Byrne and Jan Hajič and Samuel Gustman and Bhuvana Ramabhadran 8 Automatic Transcription of Czech Language Oral History in the MALACH Project: Resources and Initial Experiments Speech - automatic speech recognition 74_Josef_Psutka_abstract.txt Hynek Hermansky    Leon Rothkrantz  -The paper reports about a very interesting research project -in section 5.1 is reported that acoustic models are trained, but the training data is not mentioned and described at the end of section 5 it is concluded that the recognition rates are weak if we compare it to the results of ASR trained on Broadcast news. This is not suprisingly because the Shoah corpus is composed of spontaneous speech while the Broadcast corpus is composed of og grammatically correct text-speech recordings   LP 74_Josef_Psutka.pdf       
76 Antanas LIPEIKA 3 On Speaker Adaptation in Isolated Word Recognition Speech - automatic speech recognition 76_Antanas_LIPEIKA_abstract.txt E.G. Schukat-Talamazzini    Pavel Skrelin    Reject 76_Antanas_LIPEIKA.pdf       
77 Daniel Martins 2 Influence of text coherence’s disruption on story memorisation and interestingness Text - other 77_Daniel_Martins_abstract.txt Vladimir Petkevic  The results confirm what is crystal clear anyway - so why to perform the research? Really, the topic is definitely not interesting, there are no ideas therein. The presentation and English are both extremely bad - you should have at least spell check the text, the errors are horrible. Next time, you should devote much more time to the preparation and to the choice of the topic of the article before sending it to any kind of conference.   Yorick Wilks      Reject 77_Daniel_Martins.pdf         
78 Takafusa Kitazume 2 Automatic Video Caption Generation for Sound Containing Voice and Music Speech - speech segmentation 78_Takafusa_Kitazume_abstract.txt Taras Vintsiuk    Hynek Hermansky    Reject       
79 Juan M. Montero 2 ANESTTE: a Writer’s Assistant for a Specific Purpose Language Text - other 79_Juan_M._Montero_abstract.txt Dr. Alexander Gelboukh Kahn  Paper: ANESTTE, by Juan M. Montero et al. You should format your paper according to Springer requirements. Your paper does not present an adequate testing. Is this tool available publicly or commercially? How to get it (please give a URL)? Minor language problems:   Eneko Agirre  The authors use a template based grammar to measure some stilystic features of technical text. The aplication is interesting. It is not clear if it also adresses grammar errors. The main weaknesses are that the paper is difficult to follow, that there is no mention to other research papers in the area, or comparison to other systems (e.g. the Word grammatical . Besides de evaluation is very weak. For instance why wasn't it evaluated with technical documents vs. other kind of documents? There is no mention to language resources (e.g. dictionary) or algorithms (PoS tagging?) have been used.   Reject 79_Juan_M._Montero.pdf         
80 Robert Král 1 Word Sense Discrimination for Czech Text - word sense disambiguation 80_Robert_Kral_abstract.txt Eva Hajicova  This would be a nice piece of work if (a) formulated clearly and, first of all, understandably, (b) the approach is illustrated on more than a single example, and (c) on the basis of such a more substantive analysis the evaluation is made. The paper as it is now just makes an impression that the author has taken over an algorithm, applied it to a single ambiguous Czech word (which is not well comparable with the original tests) and has drawn conclusions from that.   Hanks Patrick    SP 80_Robert_Kral.pdf       
81 Irina Rozina 3 Interactive learning media for language, communication and culture study Dialogue - other 81_Irina_Rozina_abstract.txt Ivan Kopecek  In my opinion, this is not a research paper.   Vaclav Matousek  bad submission format, very brief presentation, I recommend to submit it to other conference or to revise this submission. I mean, you send us the contribution for other conference. Secondly - it is no full paper, it is the extended abstract.  Reject 81_Irina_Rozina.pdf         
82 Sorin Dusan 2 A System for Multimodal Language Acquisition Dialogue - user modeling 82_Sorin_Dusan_abstract.txt Elmar Noeth  I have trouble with your OOV handling. How come the computer recognizes an unknown word rather than producing a wrong hypothesis? Please elaborate Isn't the 3rd paragraph of chapter 2 and the beginning of chapter 3 the same? I'm missing experimental results - a small field test is better than none   Karel Pala  Some parts of the text on p.2, Sect.2, par.3 and p.3 Sect.3, par.1 are repeating, obviously the technique "cut and paste" was used - this can be hardly accepted. The description of the grammar appears to be quite simple and rather superficial, number of the grammar rules is really small. The semantic database mentioned in the paper does not seem to be related to any of the known knowledge represention language judging from the given Bibliography, thus it is ad hoc solution?. kp.   Reject 82_Sorin_Dusan.pdf       
84 Takeshi Akatsuka 2 Automatic Generation of Iroha-Uta Poetry Text - other 84_Takeshi_Akatsuka_abstract.txt Jaroslava Hlavacova  Not well explained. There are undefined symbols, or they are introduced later than they were used in the paper. What is "more meaningful" from the Conclusion? I can't imagine, how the using of the 2nd algorithm could be (fully) meaningful. The paper describes a japanese (sort of) game with words and even if it is interesting, I can't see its usefulness for other languages.   Eduard Hovy      Reject 84_Takeshi_Akatsuka.pdf         
85 Lim Tek Yong 2 The Exploratory of Personal Assistants Dialogue - user modeling 85_Lim_Tek_Yong_abstract.txt Ivan Kopecek    Vaclav Matousek  The contribution is nice prepared, but it deals with common (general) problems of the HCI. Therefore the relevance to TSD conference isn't high; I recommend to contribute this paper to the INTERACT 2003 conference.  Reject 85_Lim_Tek_Yong.pdf         
86 Dita Bartůšková and Radek Sedláček 2 Tools for Semi-Automatic Assignment of Czech Nouns to Declination Patterns Text - automatic morphology 86_Dita_Bartuskova_abstract.txt Steven Krauwer    Yorick Wilks  This seems an excellent, up to date (i.e. ML), approach to a clear and precise problem. I cannot judge the Czech language issues of course.  SP       
87 Tomáš Holan 1 Dependency Analyser Configurable by Measures Text - parsing and part-of-speech tagging 87_Tomas_Holan_abstract.txt Karel Oliva  *IMPROVE* your English:   Vladimir Petkevic  Nice paper with a promising content. The type of analyzer seems to be especially appropriate for languages with the high degree of word-order. I have only one critical remark: the presentation written in English contains many errors (some of them could have been corrected by means of a grammar-checker, for example!) which must definitely be corrected when you will be preparing the final version of the paper. Devote much care to the final preparation of the paper, please.   LP 87_Tomas_Holan.pdf         
88 C.K. Yang and L.J.M. Rothkrantz 2 knowledge based speech interface for handhelds Dialogue - development of dialogue strategies 88_Cheng-KeYang_abstract.txt Elmar Noeth  I like the paper very much. Nevertheless I don't see an indication of the advantage for a system designer in the paper. Why would I want to use your your system? Where do I save work? In my printout the ` and ' came out wrong.   Karel Pala  unfortunately very little is said about the handhelds in the paper, thus there is a conflict between the title and the rest of the paper and in this sense the paper is not complete. The authors should amend that, othervise the paper can be hardly accepted. kp.   LP       
89 Pavel Cenek 1 A Flexible Framework for Evaluation of New Algorithms for Dialogue Systems Dialogue - dialogue systems 89_Pavel_Cenek_abstract.txt Ivan Kopecek    Vaclav Matousek  nice prepared contribution, original topic, accept without comments.  SP 89_Pavel_Cenek.pdf       
90 Sven Hartrumpf and Hermann Helbig 2 The Generation and Use of Layer Information in Multilayered Extended Semantic Networks Text - lexical semantics and semantic networks 90_Sven_Hartrumpf_abstract.txt Yorick Wilks      Dr. Alexander Gelboukh Kahn  Paper: The Generation... by H. Helbig Needs to be formatted as Springer requires.   LP 90_Sven_Hartrumpf.pdf         
91 Zervas P. and Potamitis I. and Fakotakis N. and Kokkinakis G. 4 ON THE FIRST GREEK-TTS BASED ON FESTIVAL SPEECH SYNTHESIS: ARCHITECTURE AND COMPONENTS DESCRIPTION Speech - text-to-speech synthesis 91_Zervas_Panos_abstract.txt Genevieve Baudoin    Attila Ferencz    SP 91_Zervas_Panos.pdf         
92 Antonio Molina 3 A Hidden Markov Model Approach to Word Sense Disambiguation Text - word sense disambiguation 92_Antonio_Molina_abstract.txt Eneko Agirre  NOTE: This paper applies a previously developped method (LREC 2002) to a different test set: senseval 2 all-words. The LREC paper is not yet available but from the authors comments it can be assumed that the only difference is that of the evaluation, without further development. If that is not the case, the authors should describe better which are the improvements of their system with respect to the other publication.   Eva Hajicova    Reject 92_Antonio_Molina.pdf       
93 Aleš Horák and Vladimírr Kadlec and Pavel Smrž 3 Enhancing Best Analysis Selection and Parser Comparison Text - parsing and part-of-speech tagging 93_Ales_Horak_abstract.txt Hanks Patrick    Jaroslava Hlavacova  What is PDTB?   LP       
94 Fatiha Sadat and Masatoshi Yoshikawa and Shunsuke Uemura 3 Exploiting Thesauri and Hierarchical Categories in Cross-Language Information Retrieval Text - information retrieval 94_Fatiha_SADAT_abstract.txt Eduard Hovy      Steven Krauwer    LP 94_Fatiha_SADAT.pdf       
95 Francisco Díaz 5 Segmentation of TI-Digits Corpus with Hidden Markov Models Speech - speech segmentation 95_Francisco_Díaz_abstract.txt Nikola Pavesic    Josef Psutka  This paper does not bring any new ideas and results.   Reject 95_Francisco_Díaz.pdf         
96 Csaba Szepesvari 1 On the utility of smoothing in spoken dialogue systems Dialogue - dialogue systems 96_Csaba_Szepesvari_abstract.txt Elmar Noeth  What is r.h.s. (page V)? eg. and ie. is written as e.g. and i.e. unresolved citation on p. VI I'm sorry, but I have trouble judging what - if any - has been implemented and experimentally been verified   Karel Pala    Reject       
97 Robert Batůšek 1 An Analysis of Limited Domains for Speech Synthesis Speech - text-to-speech synthesis 97_Robert_Batusek_abstract.txt Leon Rothkrantz  -The paper is (too) short, a lot of aspects should be described in more details, i.e. in the section feature generation the authors give an indication which featues are used. But there should be a list of features. These list is used in the experiments so these details should be available. Maybe the lisit is too long but part of it should be presented. -Secton 3 starts with "all experiments have been made....", the question is which experiments? -In section 4 the authors state that the list of possible features is practically infinite. But all features are equal but some of them are more equal. So maybe it is possible to compute frequencies of the features and do the experiments with the most important features. It is also interesting to know the distribution of the feature set over the different corpora and the different testset. Maybe there will be a great variation -in table 1 some features are listed but maybe a little bit criptic for the reader who has no access to he former papers of the authors. So maybe some additional information can be provided and also some details of the Demosthenes speech synthesiser. -Maybe the authors have to consider even more restricted corpora. The corpus of dialogues of train-timetables is even more limited so maybe the featues. If we reduce the corpus maybe not all the names of railwaystations are represented but maybe all time-expressons are represented. So the frequency of the features and the relation to specific topics may be important Content-type: application/octet-stream; name="97.pdf"; type=Unknown; Content-description: 97.pdf Content-disposition: attachment Attachment converted: Gigi:97.pdf (PDF /CARO) (0000FCB9)   E.G. Schukat-Talamazzini    SP 97_Robert_Batusek.pdf         
98 Genevieve Baudoin and François Capman and Jan Černocký and Fadi El Chami and Maurice Charbit and Gérard Chollet and Dijana Petrovska-Delacrétaz 7 Advances in Very Low Bit Rate Speech Coding using Recognition and Synthesis Techniques Speech - speech coding 98_Dijana_Petrovska-Delacretaz_abstract.txt Pavel Skrelin    Taras Vintsiuk      LP 98_Dijana_Petrovska-Delacretaz.pdf       
99 Marion Mast and Thomas Ross and Henrik Schulz and Heli Harrikari 4 Different Approaches to Build Multilingual Conversational Systems Dialogue - dialogue systems 99_Marion_Mast_abstract.txt Ivan Kopecek    Vaclav Matousek  The contribution is original and from the viewpoint of the topic is carefully prepared. But, it doesn't keep the given contribution format (LNCS); it seems me, that this paper was contributed to another conference or workshop more. Therefore I recommend to take this fact into account by decision about the acceptance of this paper.  LP 99_Marion_Mast.pdf         
100 Victoria Arranz and Núria Castell and Montserrat Civit 3 Strategies to Overcome Problematic Input in a Spanish Dialogue System Dialogue - dialogue systems 100_Victoria_Arranz_abstract.txt Elmar Noeth    Karel Pala    LP 100_Victoria_Arranz.pdf         
101 Robert Hecht and Jürgen Riedler and Gerhard Backfried 3 Fitting German into N-Gram Language Models Speech - automatic speech recognition 101_Robert_Hecht_abstract.txt Hynek Hermansky    Genevieve Baudoin    LP       
102 Guy Camilleri 1 Dialogue systems and planning Dialogue - other 102_Guy_Camilleri_abstract.txt Ivan Kopecek    Vaclav Matousek  interesting contribution, I recommend to accept it for the conference  LP 102_Guy_Camilleri.pdf         
103 Kris Demuynck and Tom Laureys 2 A Comparison of Different Approaches to Automatic Speech Segmentation Speech - speech segmentation 103_Kris_Demuynck_abstract.txt Attila Ferencz    Nikola Pavesic    LP       
104 Jan Žižka and Aleš Bourek 2 Filtering of Large Numbers of Unstructured Text Documents by the Developed Tool TEA Text - information retrieval 104_Jan_Zizka_abstract.txt Vladimir Petkevic    Karel Oliva  the paper seems to be a nice description of a working system, however (at= least as to the contents of the paper), the system is hardly more than an implementation of known approaches and = techniques; hence the scientific contribution of the paper seems to be ra= ther low. I would recommend presenting the paper as a poster rather than as a full = paper.   LP 104_Jan_Zizka.pdf       
105 Stefan Grocholewski 1 Within-vowels correlation in speech and speaker recognition Speech - speaker identification and verification 105_Stefan_Grocholewski_abstract.txt Josef Psutka  To improve English.   Leon Rothkrantz  -The paper is not easy to read , there are a lot of loosely connected statements, the structure should be improved -the topic is not clear, do the authors report about speaker recognition or speaker verification? -the authors consider only vowels, but how are these vowels extracted from speech recordings, what is the error rate -it seams more natural to consider voiced as well as unvoiced parts of speech, so why do the authors restrict to vowels? no there is no proof the they got better results compared to the common methods -in the text the concepts A-B are very confusing, in Fig 1 it ids about speaker A-B in Fig 5,6 it is about databes A-B   Reject 105_Stefan_Grocholewski.pdf         
106 Yassine Ben Ayed and Dominique Fohr and Jean Paul Haton and Gérard Chollet 4 KEYWORD SPOTTING USING SUPPORT VECTOR MACHINES Speech - automatic speech recognition 106_BEN_AYED_abstract.txt E.G. Schukat-Talamazzini    Pavel Skrelin    LP       
107 Guido Aversano and Anna Esposito 2 Improved performances and automatic parameter estimation for a context-independent speech segmentation algorithm Speech - speech segmentation 107_Guido_Aversano_abstract.txt Taras Vintsiuk      Hynek Hermansky    LP 107_Guido_Aversano.pdf       
108 Marek Labuzek 2 English Translator - A Bi-directional Polish-English Translation System Text - machine translation 108_Marek_Labuzek_abstract.txt Vladimir Petkevic  The task of machine translation is an extremely difficult. All modules must be next to perfect to perform the task in a relatively satisfactory way. Your description is very general and the whole approach is very naive. For instance, if your POS tagger within analysis of the input language will be bad you can hardly get good results on output. If the quality of the parser is, as you say, very unsatisfactory, why do you describe the whole system because it simply can't work. Some results are presented but they are very bad (eg. 86 % tagger accuracy). The whole method is generally relatively sound but the individual components seem to be designed in a very naive way. For Polish, parsing must be performed with VERY DEEP linguistic insights into the structure of such a syntactically complicated language as Polish.   Yorick Wilks      Reject 108_Marek_Labuzek.pdf       
109 Zdenek Svoboda 1 The Encyclopedia Expert Text - information retrieval 109_Zdenek_Svoboda_abstract.txt Dr. Alexander Gelboukh Kahn  Paper: "The Encyclopedia Expert" by Zdenek Svoboda. Format: remove page numbers. Abstract: 1-2 more lines of details on your method (use of XML, etc.). Give the reader an idea of why it is worth effort to read your paper.   Eneko Agirre  This paper presents a Q/A system developped following some well-known ideas. It makes more for a demo than a research paper. It does not present innovative work and the linguistic modelling is limited to pattern matching.   Reject 109_Zdenek_Svoboda.pdf         
110 Pascal Nocera and Georges Linares and Dominique Massonié and Loic Lefort 4 Phoneme Lattice Based A* Search Algorithm for Speech Recognition Speech - automatic speech recognition 110_Pascal_NOCERA_abstract.txt Genevieve Baudoin    Attila Ferencz    LP       
111 Christophe Heintz 1 Naming and the management of social interactions Dialogue - other 111_Christophe_Heintz_abstract.txt Elmar Noeth  Interesting paper I feel that it is not clear what kind of human-machine interaction - if any - you have in mind: human-machine interaction to support a theory of human learning, a conversational smalltalk agent that will learn meaning to pass a Turing test (what for?) or practical real-life speech applications? Each of these goals has a right to exist, but I would like to know which one you mean.  Karel Pala  Bad English, grammatical errors in agreement: The first sentence in the Introduction: "Development ... lead(s?) to think ... 6th line in Introduction: "word comprehension and production are handle(d?)... The paper is just an interesting essay, however, it does not seem to offer any applicable results that could be somehow exploited in the field of NLP. In my opinion, the paper is not suitable for TSD Conference. kp.   Reject       
112 P. Matějka and P. Schwarz and M. Karafiát and J. Černocký 4 Some like it Gaussian ... Speech - automatic speech recognition 112_Pavel_Matejka_abstract.txt Nikola Pavesic    Josef Psutka  Nice paper. But, it is not clear whether all experiments were performed on the HTK toolkit or some other software tool? Also memory requirements and processing time necessary for common technique (baseline system) in comparison with technique based on gaussianization could be mentioned.   SP 112_Pavel_Matejka.pdf       
113 Dominic Widdows and Scott Cederberg and Beate Dorow 3 Visualisation Techniques for Analysing Meaning Text - lexical semantics and semantic networks 113_Dominic_Widdows_abstract.txt Eva Hajicova  The tools you describe are interesting and useful, especially for showing important connections that would not be seen from the data as such. A graphical remark: the example in Secti 3 should certainly be oplace in some other position in the text. It is also not convincing enough if you say that the top neighbours of the word could be determined by the users - will they have some tool for that? A list of possibilities? A mssing verb in the second sentence of the paragraph starying with: The  Hanks Patrick    LP       
114 Aldezabal 5 Learning Argument/Adjanct distinction for Basque Verbs Text - other 114_Aldezabal_abstract.txt Jaroslava Hlavacova    Eduard Hovy      will not arrive       
115 Liang Huang and Yinan Peng and Huan Wang and Zhenyu Wu 3 Part-of-Speech Tagging for Old Chinese Text - parsing and part-of-speech tagging 115_Liang_HUANG_abstract.txt Steven Krauwer  very original and interesting!   Karel Oliva  The contribution of the paper is to be found - apart from dealing with Cl= assical Chinese - mainly in two areas 1. in proposing a tagset for Classical Chinese 2. in developing a statistical tagger for Classical Chinese. Apart form that, you also report creating a small tagged corpus of Classi= cal Chinese. To start with the last, I am slightly afraid that a training corpus of 1.= 000 words (i.e. about 50 sentences) is really TOO small to provide any si= gnificant results. The same holds for your test corpus - 200 words, i.e. about 10 sentences. I dare say that no really representative results can be achieved with thi= s size of corpora. Aport form that, you say you "present simple-yet-effective methods to han= dle the problems (which occur due to the difference of Classical Chinese = from Indoeropean languages)". These methods, however, as a rule turn to be nothing novel but just simpl= ification of the standardly used techniques (e.g., you use simple bigrams= instead of smoothing trigrams). In this respect, it seems that the paper poses interesting problems but d= oes not present clear answers, neither in theory nor in technology, For these reasons, I would propose the paper (if accepted) to be presente= d rather as a poster than as a full paper.   LP 115_Liang_HUANG.pdf         
116 Gregory Martynenko 1 Statistical Taxonomization of Literary Corpus Text - text corpora 116_Gregory_Martynenko_abstract.txt Karel Oliva  It would be nice to see some more evaluation - in the sense of numbers. E= .g., it would be nice to see in numbers (i.e. QUANTITATIVELY) why some au= thors are nearer to each other than others. Just add some tables or so. *IMPROVE* your English for the publication (if accepted): - generally, commata (you put them where they would stay in Russian, but = English has different rules)  Vladimir Petkevic  The article seems to be interesting but too general. It would be adequate to depict some of the techniques in a more detailed way. However, I am not sure that the topic is crucially relevant for the conference. Some (statictical) results and techniques and at least a slightly (I know that the limit is 8 pages at most!) deeper analysis of a certain literary period should also have been shown instead of 2 pages consisting of significant Russian adjectives and nouns.   Reject       
117 Marina Lublinskaya and Tatiana Sherstinova 2 Audio Collections of Endangered Arctic Languages in the Russian Federation Speech - other 117_Tatiana_Sherstinova_abstract.txt Leon Rothkrantz  -the paper is about an interesting problem -the paper is written as a report/essay, no technical details are reported, it is difficult to repeat the study in similar cases -the authors state that they solved a lot of technical problems but not how -the authors report that the studied languages have no official spelling, but how did they solve that problem -the paper has some spelling errors: section 2, line 1 Nenets is the small group of people section 2, third paragraph bothe Nenets and Nganasa languages belong section 2 fourth paragraph their children to have better life conditions section 2 fourth paragraph in most of the cases The Netherlands on the Web section 5 fourth paragraph The speech corpus section 6 line 12 Now they are students   E.G. Schukat-Talamazzini    LP       
118 Junko Araki 4 Action Vectors and Its Application to Interactive Dialogue Systems Dialogue - development of dialogue strategies 118_Junko_Araki_abstract.txt Ivan Kopecek    Vaclav Matousek  The paper describes interesting application, but it doesn't keep the given paper format (LNCS). I recommend to accept the paper as the poster, but after the revision.  Reject 118_Junko_Araki.pdf       
119 Ewa Lukasik 1 Elements of speaker variability in some voiceless phonemes Speech - speaker identification and verification 119_Ewa_Lukasik_abstract.txt Pavel Skrelin    Taras Vintsiuk    Reject 119_Ewa_Lukasik.pdf         
120 Igor A. Bolshakov 2 Word Combinations as an Important Part of Modern Electronic Dictionaries Text - lexical semantics and semantic networks 120_Igor_A._Bolshakov_abstract.txt Ivan Kopecek    Eneko Agirre  The goal of the paper is not clearly stated. The writing is rather awkward, making use of the special terminology used by Melcuk. If I understood the main point is to argue that some hand labor is required to code idioms, terms and collocations in a lexicon. A fact that nobody can argue. It seems that the authors also claim that it is a feasible task, but no hard data is provided.   Reject       
121 Alexander Gelbukh 2 A Method for Development of Automatic Morphological Analysis Systems for Inflective Languages Text - automatic morphology 121_Alexander_Gelbukh_abstract.txt Eneko Agirre  Interesting method for avoiding problematic morphological analysis. Unfortunately there is no evaluation of the method: does it work 100% correct for known words? 100% for unknown words? How does it compare to other systems for russian? Other comments: - section 2: an example of stem alternation should be provided at the beginning - section 2.5: the example for stopping is not very clear. In step 3 you check the potential stem stopp in the dictionary. Is stopp in the dictionary? Is stopp one of those stem alternations generated in the generation phase?   Eva Hajicova  The paper makes an impression that the authros are not acquainted with the more recent literature on the subject matter, and in addition, they do not make the right sense of the literature they do know (at least that they quote). Analysis by synthesis make be 'recent' in AI literature, but is very traditional n computational linguistics. In the domai the authors want to apply this approach, I am afraid that it might bring an undesirable increase of processing time. Remark to point 2 in Sect. 2.5: there may be more than one set of grammemes for the mentioned inflectional ending (-ing: writing may be a noun or a verb, or at least two verbal forms: nominalization and verbal participle) .One positive thing is the extensive use of the most invaluable Zaliznjak's dictionary.   Reject       
122 Rodolfo A. Pazos R. and Alexander Gelbukh and J. Javier González B. and Erika Alarcón R. and Alejandro Mendoza M. and A. Patricia Domírnguez S 6 Spanish Natural Language Interface for a Relational Database Querying System Text - other 122_Rodolfo_A._Pazos_R._abstract.txt Hanks Patrick    Jaroslava Hlavacova    LP       
123 Marek Veber 1 Formal system for collocations in Czech Text - automatic morphology 110_Marek_Veber_abstract.txt Eduard Hovy      Steven Krauwer  I don't think one should embark on any activity with respect to collocations without having looked at standard sources such as Melcuk and Pustejovsky.   Reject 110_Marek_Veber.pdf         
124 Diana Zaiu Inkpen 2 Automatic Sense Disambiguation of the Near-Synonyms in a Dictionary Entry Text - word sense disambiguation 124_Diana_Zaiu_Inkpen_abstract.txt Eneko Agirre  A paper on WSD using a set of well-known unsupervised techniques. The only novelty is to use a small sample of training data to choose the best combination. - section 2: in general it is not always clear when it is the near-synonymy entry, and when the WordNet entry. - the last sentence of the first paragraph in section 2.1 does not add anything. - 2.3: it is not clear what is it that you intersect - 2.6: when talking about the 904 datapoints a reference should be made to section 4 - 3: there is no mention to what is accuracy, and which is the coverage. this point is clarified later in section 6. - 3: I think that it should be stressed that you measure each sense independently. This makes comparison difficult to systems which measure accuracy in the Senseval way. - 3: do all algorithms return always an answer for each sense? - 4: data on the ambiguity on the gold standard (around 450 senses for 282 near-synonyms) - 5: the comparison with the Lesk algorithm is not fair, the simple lesk algorithm gets very low results. - 5: the way to calculate precision and coverage in Senseval is different. refer to their web page.   Karel Oliva    Reject       
127 Schwarz Jana 2 Dialogue Models for Bilingual Human-Computer Interaction in a City Information System Dialogue - development of dialogue strategies 127_Schwarz_Jana_abstract.txt Elmar Noeth  I feel it would be better for most of the readers if the example in chapter 6 is presented translated into English Please provide a reference for the GAT system (chapter 1) also: do you have any experimental evidence for the interesting claim that people expect the system to retain char. of human agents (same paragraph) there is quite a few typos: develope, withal, detailled, exhausting instead of exhaustive templats, Additionaly If you have the chance, have somebody, who is very competent in English, proof-read the final version   Karel Pala  -- kp.   Reject       
128 Elena Karagjosova and Ivana Kruijff-Korbayová 2 An Analysis of Conditional Responses in dialogue Speech - other 128_Elena_Karagjosova_abstract.txt Elmar Noeth  Example 2:8 is missing   Vaclav Matousek  Interesting contribution, but the problem could be described more comprehensive (why not 8 pages ?). I recommend to extend the papers and then accept for the conference.  SP       
130 Ghuilaine Clouet 1 Etude de la perception de la qualite des sites Web par les usagers en vue de dimensions d'evaluation et resultats utiles a la conception Text - other 130_Ghuilaine_Clouet_abstract.txt Genevieve Baudoin          Reject 130_Ghuilaine_Clouet.pdf         
131 Fernando LLopis Pascul 1 Passage Selection to Improve Question Answering Text - information retrieval 131_Fernando_LLopis_Pascul_abstract.txt Karel Pala  The text of the paper contains grammatical errors (agreement) "...a query term appear(s?) in each document ..." The explanations in the paper seem to be not quite clearly formulated, e.g. the authors refer to ATT IR system but I was not able to decipher if they have in mind their own system and how it may be related to IR n system mentioned in the first sect. Introduction. In sect. 3.1 Porter stemmer is mentioned but there's no reference to it in the Bibliography - thus it is impossible to even guess the reliability some of their results. kp.   Ivan Kopecek    Reject 131_Fernando_LLopis_Pascul.pdf       
132 Karel Pala 2 A Procedure for the Semiautomatic Building of Consistent Dictionary Definitions Text - other 132_Karel_Pala_abstract.txt Eneko Agirre  Interesting work. The relation of the goals of the paper and the experiments is not very clear. I think the goals of the paper should be better stated to reflect the experiments. For instance: - is the goal of the automatic analysis of dictionary definitions to enrich WN? - is the goal of the analysis to construct proper definitions for WN? - is it a paper on the relation between human definitions and relations in LKBs? - There is a lack to references in the area of automatic analysis of definitions in dictionaries (Amsler, 81; Vossen, 89; Agirre et. al, 2000 in Euralex for a recent paper with more references). - section 3: I don't think the data confirms that the definitions give always a genus. Some of the most frequent head nouns are who, sort, part. In the examples in the 4th page we can find "a piece of, a set of, a person who". These have been traditionally taken as specific relators in opposition to genus+differentia definitions. - section 3: what is the meaning of 1st file in figure 1. - section 3: is there any measure of evaluation of the results of the syntactic analysis (accuracy, coverage) - Finally some sentences are too long, and somewhat awkward to understand.   Eduard Hovy      Reject       
133 Elena Boian 1 The lexical-morphological analysis system of the Roumanian language Text - automatic morpholgy 133_abstract.txt Ivan Kopecek    Karel Pala    Reject       
134 Agnieszka Mykowiecka 7 A Large-Scale Corpus of Polish and Tools for its Annotation Text - text corpora 134_Agnieszka_Mykowiecka_abstract.txt             Demonstration          
135 Antanas Lipeika 3 ISOLATED WORD RECOGNITION AND VISUALIZATION SOFTWARE Speech - automatic speech recognition 135_Antanas_Lipeika_abstract.txt             Demonstration          
136 Bronius Tamulynas 2 Computer-based Translation from English to Lithuanian Text - machine translation 136_Bronius_Tamulynas_abstract.txt             Demonstration          
137 Alexander Troussov 2 IBM Dictionary and Linguistic Tools system “Frost” Text - other 137_Alexander_Troussov_abstract.txt             Demonstration          
138 Piotr Banski 1 XML architecture for a modern corpus Text - text corpora 138_Piotr_Banski_abstract.txt             Demonstration          
139 Qian Hu 5 The MITRE Audio Hot Spotting Prototype - Using Multiple Speech and Natural Language Processing Technologies Speech - other 139_Qian_Hu_abstract.txt             Demonstration          
141 Pekar Darko pekard@eunet.yu 3 ALFANUM SYSTEM FOR CONTINUOUS SPEECH RECOGNITION Speech - automatic speech recognition 141_Pekar_Darko_abstract.txt             Demonstration          
142 Elena Boian 2 Romanian words inflection Text - automatic morphology 142_Elena_Boian_abstract.txt             Demonstration          
143 Elena Karagjosova,, korbay@CoLi.Uni-SB.DE 3 GoDiS - Issue-based dialogue management in a multi-domain, multi-language dialogue system Dialogue - dialogue systems 143_Elena_Karagjosova_abstract.txt             Demonstration          
144 Demidova Valentina 2 Hyphenation algorithm for Romanian language words Text - parsing and part-of-speech tagging 144_Demidova_Valentina_abstract.txt             Demonstration          
145 Styve Jaumotte 3 Semantic Knowledge in an Information Retrieval System Text - information retrieval 145_Styve_Jaumotte_abstract.txt             Demonstration          
146 Petr Schwarz 4 Keyword spotting system Speech - automatic speech recognition 146_Petr_Schwarz_abstract.txt             Demonstration          
201 Michael Bodasov 2 Generator module for InBASE NL data base Interface system Text - information retrieval 201_Michael_Bodasov_abstract.txt             Demonstration          
202 Beate Dorow 3 Visualisation Techniques for Analysing Meaning Text - knowledge representation and reasoning 202_Beate_Dorow_abstract.txt             Demonstration          
203 Zhiping Zheng 1 Deploying Web-based Question Answering System to Local Archive Text - information retrieval 203_Zhiping_Zheng_abstract.txt             Demonstration          
204 Pavel Rychly 1 Advance concordances with Bonito Text - text corpora 204_Pavel_Rychly_abstract.txt             Demonstration          
205 Jan Šedivý 1 Demonstration of multi-modal applications on IPAQ Speech - automatic speech recognition 205_Jan_Šedivý_abstract.txt             Demonstration          
206 pics pics pics Text - text corpora 206_pics_abstract.txt