Báo cáo khoa học: "A Cross-Lingual ILP Solution to Zero Anaphora Resolution" potx

... Association for Computational LinguisticsA Cross-Lingual ILP Solution to Zero Anaphora ResolutionRyu IidaTokyo Institute of Technology2-12-1,Ôokayama, Meguro,Tokyo 152-8552, Japanryu-i@cl.cs.titech.ac.jpMassimo ... (7)3.2 A subject detection modelThe greatest difficulty in zero anaphora resolutionin comparison to, say, pronoun resolution, is zero anaphora detection. Simply relying for this on theparser ... accurate esti-mation.4.5 Results with zero anaphora onlyIn zero anaphora resolution, we need to find all pred-icates that have anaphoric unrealized subjects (i.e. zero pronouns which have an antecedent...

Báo cáo khoa học: "A Syntax-Free Approach to Japanese Sentence Compression" potx

... Syntax-Free Approach to Japanese Sentence CompressionTsutomu HIRAO, Jun SUZUKI and Hideki ISOZAKINTT Communication Science Laboratories, NTT Corp.2-4 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0237 ... summariza-tion systems often have to process megabytes ofdocuments. Parsers are still slow and users of on-827demand summarization systems are not prepared to wait for parsing to ﬁnish.3 A Syntax Free ... evaluation,i.e., we presented the compressed sentences to sixhuman subjects and asked them to evaluate thesentence for ﬂuency and importance on a scale 1(worst) to 5 (best). For each source sentence, theorder...

Tài liệu Báo cáo khoa học: "A Ranking-based Approach to Word Reordering for Statistical Machine Translation" doc

... converted to dependency trees us-ing Stanford Parser (Marneffe et al., 2006). We con-vert the tokens in training data to lower case, andre-tokenize the sentences using the same tokenizerfrom ... sensitive to parser er-rors; on the other hand, integrated model is forced to use a longer distortion limit which leads to moresearch errors during decoding time. It is possible to 918use ... work, we plan to extend the rankingmodel to handle reordering between multiple lev-els of source trees. We also expect to explore bet-ter way to integrate ranking reorder model into SMTsystem...

Tài liệu Báo cáo khoa học: "An Equivalent Pseudoword Solution to Chinese Word Sense Disambiguation" ppt

... methods lies in the knowledge acquisition solutions they adopt. 2.1 Automatic Generation of Training Corpus Automatic corpus tagging is a solution to WSD, which generates large-scale corpus ... EPs. A Chinese thesaurus is adopted and revised to meet this de-mand. Extended Version of TongYiCiCiLin To extend the TongYiCiCiLin (Cilin) to hold more words, several linguistic resources ... word, which is called an atom word group, an atom class or an atom node. The words in the same atom node hold the smallest semantic dis-tance. From the root node to the leaf node, the sense...

Tài liệu Báo cáo khoa học: "A Fully Bayesian Approach to Unsupervised Part-of-Speech Tagging∗" docx

... differences hold to alesser degree when a partial dictionary is provided.With MLHMM, different tokens of the same wordtype are usually assigned to the same cluster, buttypes are assigned to clusters ... t. We willuse τ and ω to refer to the entire transition and out-put parameter sets. This model assumes that theprior over state transitions is the same for all his-tories, and the prior over ... exchangeability iscrucial to the inference algorithm we describe next.2.3 Inference To perform inference in our model, we use Gibbssampling (Geman and Geman, 1984), a stochasticprocedure that...

Tài liệu Báo cáo khoa học: "A Feature Based Approach to Leveraging Context for Classifying Newsgroup Style Discussion Segments" pptx

... state of a simple finite-state automaton that only has two states. The automaton is set to initial state (q0) at the top of a message. It makes a transition to state (q1) when it encounters ... is to enable the quality and nature of discussions that occur within an on-line discussion board to be communicated in a summary to a potential new-comer or group moderators. We propose to ... where a message appears. We refer to this feature as deep. This is expected to improve per-formance to the extent that thread initial messages may be rhetorically distinct from messages that...

Tài liệu Báo cáo khoa học: "A Limited-Domain English to Japanese Medical Speech Translator Built Using REGULUS 2" doc

... on the other. We propose to demon-strate a prototype system instantiating thisarchitecture, which has been built on topof the Open Source REGULUS 2 platform.The prototype translates spoken ... RoadMountain View, CA 94040vvandal3@aol.comHitoshi Isahara, Kyoko KanzakiCommunications Research Laboratory3-5 HikaridaiSeika-cho, Soraku-gunKyoto, Japan 619-0289{isahara,kanzaki}@crl.go.jpBeth ... yes-noquestions about headache symptoms fromEnglish to Japanese, using a vocabulary ofabout 200 words.1 Introduction and motivationLanguage is crucial to medical diagnosis. Dur-ing the initial...

Báo cáo khoa học: "A Nonparametric Bayesian Approach to Acoustic Model Discovery" docx

... thefuture, we plan to explore phonological context anduse more ﬂexible topological structures to modelacoustic units within our framework.AcknowledgementsThe authors would like to thank Hung-an ... R39 to denote the tthfeature frame of theithutterance. Fig. 1 illustrates how the speech signalof a single word utterance banana is converted to asequence of feature vectors xi1 to xi11.Boundary ... gi0 to be the beginning ofan utterance, and Li to be the number of boundaryframes in an utterance. For the example shown inFig. 1, Liis equal to 6.Segment (pij,k) We deﬁne a segment to...

Báo cáo khoa học: "A Two-step Approach to Sentence Compression of Spoken Utterances" pdf

... first step, 8 anno-tators were asked to select words to be removed to compress the sentences. In the second step, 6 an-notators (different from the first step) were asked to pick the best one ... andreranking is able to yield additional gain, espe-cially when training is performed to take intoaccount multiple references.1 IntroductionSentence compression aims to preserve the most ... In this work, we propose to use discriminative training to rerank the candi-dates generated in the first step. Reranking has beenused in many tasks to find better global solutions,such as machine...

Báo cáo khoa học: " a Movie Dialogue Corpus for Research and Development" potx

... of the resulting dialogue collection. Total number of scripts collected 911 Total number of scripts processed 753 Total number of dialogues 132,229 Total number of speaker turns 764,146 Average ... inserted within the turn, misspelling of speaker names, etc. In addition to this, a semi-automatic process was still necessary to filter out movie scripts exhibiting extremely different layouts ... dia-logue systems can be grouped into two major cate-gories: task-oriented and chat-oriented systems. In the first case, the system is required to help the user to accomplish a specific goal or...

Xem thêm