Tài liệu Báo cáo khoa học: "Unsupervised Topic Modelling for Multi-Party Spoken Discourse" ppt

... Association for Computational LinguisticsUnsupervised Topic Modelling for Multi-Party Spoken DiscourseMatthew PurverCSLIStanford UniversityStanford, CA 94305, USAmpurver@stanford.eduKonrad ... simultaneously address the prob-lems of topic segmentation and topic identiﬁcation: automatically segmenting multi-party meetings into topically co-herent segments with performance whichcompares well ... drawing the topic assignment for eachword, zu,i, conditioned on all other topic assign-ments, z−(u,i), all topic change indicators, c, andall words, w; and then drawing the topic changeindicator...

Tài liệu Báo cáo khoa học: "Unsupervised Translation Induction for Chinese Abbreviations using Monolingual Corpora" ppt

... approach is to transform the abbreviation425into its full-form for which the current SMT systemknows how to translate. For example, if the baselinesystem knows that the translation for “ ” is“Hong ... Performance4.5.1 Precision on Translations of ChineseFull-form Phrases For the relations manually tagged as correct inSection 4.4, we manually look at the top-5 transla-tions for the full-form ... Translation Induction for ChineseAbbreviationsGiven a Chinese abbreviation and its full-form, weinduce English translation entries for the abbrevia-tion by using the full-form as a bridge. Speciﬁcally,we...

Tài liệu Báo cáo khoa học: "Bilingual Sense Similarity for Statistical Machine Translation" ppt

... co-occurrence counts of the two units. Therefore, questions emerge: how good is the sense similarity computed via VSM for two units from parallel corpora? Is it useful for multi-lingual applications, ... (WSD) techniques in SMT for translation selection. However, WSD techniques for SMT do so indirectly, using source-side context to help select a particular translation for a source rule. source ... the definition in section 3 from the full training data for α and for γ, re-spectively. coocfCandcooceC are the contexts for α and γ when α and γco-occur. Obviously,...

Tài liệu Báo cáo khoa học: " A Declarative Language for Implementing Dynamic Programs∗" pptx

... acquire more over time: we in-tend for it to generalize and encapsulate best practices,and serve as a testbed for new practices. Dyna is now be-ing used for parsing, machine translation, morphologicalanalysis, ... probabilities, whichpermits heuristic early stopping before the agenda isempty. With viterbi values, it amounts to uniform-costsearch for the best parse, and an item’s value is guaran-teed ... Parameter TrainingDyna provides facilities for training parameters. For ex-ample, from Fig. 1, it automatically derives the inside-outside (EM) algorithm for training PCFGs.How is this possible?...

Tài liệu Báo cáo khoa học: "Using Confidence Bands for Parallel Texts Alignment" pptx

... longer afford to waste human time and effortbuilding manually these ever changing and in-complete databases or design language specificapplications to solve this problem. The need for an automatic ... brackets). For average size texts (e.g. the Written Ques-tions), these words account for about 5% of thetotal (about 3k words / text). This number variesaccording to language similarity. For instance,on ... Y-axis intercept (thevalue of y when x is 0), substituting x for thePortuguese word position. For Table 3, the ex-pected word position for the word I at pt wordposition 3877 is 0.9165 × 3877 +...

Tài liệu Báo cáo khoa học: "AN INDEXING TECHNIQUE FOR IMPLEMENTING COMMAND RELATIONS" ppt

... requierments for the formalism used. Hence, the indexing technique has a wide spectrum of applications for testing command relations in syntactic analysis. Futhermore, this method can also be used for ... special cases of definitions 1.1 and 1.2 for the property of being a set of maximal projections. Before I formulate the general command definition in a formal way, I will now quote some other ... between syntactic and semantic structure, since the necessary syntactic information is encoded in indices. Therefore, this information can be moved to any required position in the semantic structure...

Tài liệu Báo cáo khoa học: "Translation Model Adaptation for Statistical Machine Translation with Monolingual Topic Information" doc

... context in-formation at the sentence level, we adopt thetopical context information in our method for the following reasons: (1) the topic informa-tion captures the context information beyondthe ... spe-ciﬁc topical contexts, and the topical context infor-mation has a great effect on translation selection. For example, “bank” often occurs in the sentencesrelated to the economy topic when ... the topical context in-formation is integrated into the posterior prob-ability distribution, avoiding the sparseness ofword or POS features; (3) the topical contextinformation allows for...

Tài liệu Báo cáo khoa học: "A Statistical Model for Unsupervised and Semi-supervised Transliteration Mining" pptx

... labelled information for training. Our sys-tem extracts transliteration pairs in an unsupervisedfashion. It is also able to utilize labelled informationif available, obtaining improved performance.We ... of the Association for Computational Linguistics, pages 469–477,Jeju, Republic of Korea, 8-14 July 2012.c2012 Association for Computational LinguisticsA Statistical Model for Unsupervised ... alignment-basedstring similarity for transliteration mining. Thesemethods model the conditional probability distribu-tion and require supervised/semi-supervised infor-mation for learning. We propose...

Tài liệu Báo cáo khoa học: "Unsupervised Semantic Role Induction with Global Role Ordering" doc

... PARAMETERS —————– for all predicate p do for all voice vc ∈ {active, passive} dodraw θorderp,vc∼ Dirichlet(αorder) for all interval I dodraw θSRp,I∼ Dirichlet(αSR) for all adjacency ... the ordering of core roles is importantinformation for SRL systems, and (ii) the intervalsbounded by core roles provide good context infor-mation for classiﬁcation of other roles.4 ConclusionsWe ... 83.72 78.08 80.80Table 2: Performance variation with the number ofPRs (excluding ST ART , END and P RED)With only this additional ordering information, theperformance is the same as the baseline....

Tài liệu Báo cáo khoa học: "Unsupervised Discourse Segmentation of Documents with Inherently Parallel Structure" pdf

... part-specific topic proportions β(k)∼GEM(α(k)) for k ∈ {1, . . . , K}.• Choose the part-specific language models φ(k)i∼Dir(γ(k)) for k ∈ {1, . . . , K} and i ∈ {1, 2, . . .}.• For each ... parts.The formal definition of our model is as follows:• Draw the document-level topic proportions β(doc)∼GEM(α(doc)).• Choose the document-level language model φ(doc)i∼Dir(γ(doc)) for ... time using non-informativehyperpriors (as we do in our experiments), or setmanually to indicate preferences of segmentationgranularity.At inference time, we enforce each latent topic z(k)nto...

Xem thêm