Báo cáo khoa học: "Learning Semantic Correspondences with Less Supervision" potx

... pages 91–99,Suntec, Singapore, 2-7 August 2009.c2009 ACL and AFNLPLearning Semantic Correspondences with Less SupervisionPercy LiangUC Berkeleypliang@cs.berkeley.eduMichael I. JordanUC ... grounded language acqui-sition is learning the correspondences between arich world state and a stream of text which refer-ences that world state. To deal with the high de-gree of ambiguity present ... world state s, we propose a generative modelp(w | s) with latent variables specifying this cor-respondence.Our model combines segmentation with align-ment. The segmentation aspect of our model...

Tài liệu Báo cáo khoa học: "Learning to Translate with Multiple Objectives" doc

... toBLEU (BLEU=.265,RIBES=.665) vs. anotherweight with higher RIBES but poorer BLEU,e.g. (.255,.675). Nevertheless, both the PMOand Linear-Combination with various (p1, p2)samples this joint-objective ... different metrics Mk(h) forevaluating the quality of h. Without loss of gen-erality, we assume metric scores are bounded be-tween 0 and 1, with 1 being perfect. Each hypoth-esis h can be mapped ... circle(o) are pareto-optimal, while those with plus (+) arenot. To visualize this, take for instance the pareto-optimal point (0.4,0.7). There is no other point with either (metric1 > 0.4 and metric2...

Tài liệu Báo cáo khoa học: "Learning Word Senses With Feature Selection and Order Identiﬁcation Capabilities" pdf

... reduced to 100 dimensions us-ing SVD. If the dimension of context vector was less than 100, all of latent semantic vectors with non-zero eigenvalue were used for subsequent clus-tering. We also tested ... senses for this verb and adjective,but slightly less reliable for this noun. Compared with their conclusion, we can ﬁnd that our resultis consistent with it for “hard”. But there is somedifferences ... parameterset is chosen as an initial condition for EM iteration with K − 1 subclasses. This operation will avoid acomplete minimization with respect to π, µ, and Rfor each value of K.Table 2: Four...

Báo cáo khoa học: "Learning Script Knowledge with Web Experiments" doc

... MSA to prefer the alignmentof two phrases if they are semantically similar, i.e.it should cost more to align ‘exit’ with ‘eat’ than‘exit’ with ‘leave’. Thus we take a measure of se-mantic (dis)similarity ... scenarios, there were sig-niﬁcant numbers of ESDs both with the minimumlength of 5 and the maximum length of 16 and ev-erything in between. Combined with the fact that93% of all individual event descriptions ... granularity, withoutdetermining the exact number of clusters before-hand. We optimized this parameter automaticallyfor each scenario: The system picks the value thatyields the optimal result with...

Báo cáo khoa học: "Learning to Translate with Source and Target Syntax" pot

... bijec-tively with those of α.• The terminal-labeled frontier nodes of γ arealigned (many-to-many) with those of α.In the substitution operation, an aligned pair ofsubstitution nodes is rewritten with ... previous work (Chiang, 2007), with differences shown in Table 2. Rules with non-terminals were extracted from a subset of the data(labeled “Core” in Table 2), and rules without non-terminals were ... virtual nodes (see Fig-ure 3b): for every elementary tree sequence with roots X1, . . . , Xn, create a new root node with a.. NP..... NNP↓..... NNP↓..... NN... Minister......

Báo cáo khoa học: "Reducing semantic drift with bagging and distributional similarity" pdf

... very poorly with others, making comparisons unreli-able. We exploit this wide variation with bagging, sampling from automatically ex-tracted seeds to reduce semantic drift.However, semantic drift ... ﬁltered to remove pat-terns appearing with less than 7 terms4. The statis-tics of the resulting dataset are shown in Table 1.3.2 Semantic CategoriesThe semantic categories we extract from ... been developedto reduce semantic drift.1One approach is to extract multiple semantic categories simultaneously, where the individualbootstrapping instances compete with one anotherin an attempt...

Báo cáo khoa học: "Learning Semantic Links from a Corpus of Parallel Temporal and Causal Relations" doc

... syntactic features from Section 4. Semantic The semantic features from Section 4.All Both syntactic and semantic features.All+Tmp (Causals Only) Syntactic and semantic features, plus the gold-standard ... returned-waiting with BEFORE since returned occurred ﬁrst, and acausal classiﬁer should label it CAUSAL since thisand can be paraphrased as and as a result.We identiﬁed both syntactic and semantic ... Fortemporal relations, the F-measures of all SVM mod-els exceeded all baselines, with the combination ofsyntactic and semantic features performing 5 pointsbetter (43.6% precision and 55.8% recall)...

Báo cáo khoa học: "Learning Semantic Categories from Clickthrough Logs" pdf

... of learning semantic categories, all the previous studies usingweb search logs rely on web search query logs.Therefore, we propose to use web searchclickthrough logs to learn semantic categories.Joachims ... dataset) were provided to WSCD09: Work-through logs to improve semantic category acqui-sition in both precision and recall.We cast semantic category acquisition fromsearch logs as the task of ... pattern which co-occurs with terms other thanthe Travel domain such as “google” and “yahoo.”Since the term “china” shares two patterns, “♯ air-lines” and “♯ map,” with “singapore,” the label...

Báo cáo khoa học: "Learning Common Grammar from Multilingual Corpus" potx

... word orderin most European languages is subject-verb-object(SVO), and some words with similar forms areused with similar meanings in different languages.The reasons for these common properties ... grammatical category clustered acrosslanguages as well as within a language. Fig-ure 3 shows examples of inferred common gram-mar rules with high probabilities. Grammar rulesthat seem to be common ... (2)where C(r; z, l, d) is the count of rule r that oc-curs in the dth sentence of language l with parsetree z. The multinomial weights are calculated asfollows,πθlAt= expEq(θlA)log...

Báo cáo khoa học: "Unsupervised Word Alignment with Arbitrary Features" potx

... J. DeNero, andD. Klein. 2010. Painless unsupervised learning with features. In Proc. of NAACL.P. Blunsom and T. Cohn. 2006. Discriminative wordalignment with conditional random ﬁelds. In Proc. ... jump conjoined with an indicator feature for the target length n, the dis-cretized jump feature conjoined with the class of saj,and the discretized jump feature conjoined with theclass ... conclude with an analysis of relatedprior work (§6) and a general discussion (§8).2 ModelIn this section, we develop a conditional modelp(t | s) that, given a source language sentence s with length...

Xem thêm