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

... pages 91–99, Suntec, Singapore, 2-7 August 2009. c 2009 ACL and AFNLP Learning Semantic Correspondences with Less Supervision Percy Liang UC Berkeley pliang@cs.berkeley.edu Michael I. Jordan UC ... grounded language acqui- sition is learning the correspondences between a rich world state and a stream of text which refer- ences that world state. To deal with the high de- gree of a...

Ngày tải lên: 17/03/2014, 01:20

... to BLEU (BLEU=.265,RIBES=.665) vs. another weight with higher RIBES but poorer BLEU, e.g. (.255,.675). Nevertheless, both the PMO and Linear-Combination with various (p 1 , p 2 ) samples this joint-objective ... different metrics M k (h) for evaluating 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...

Ngày tải lên: 19/02/2014, 19:20

... 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 result is consistent with it for “hard”. But t...

Ngày tải lên: 20/02/2014, 16:20

... MSA to prefer the alignment of 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 minimum length of 5 and the maximum length of 16 and ev- erything in between. Combined with the fact that 93%...

Ngày tải lên: 07/03/2014, 22:20

... bijec- tively with those of α. • The terminal-labeled frontier nodes of γ are aligned (many-to-many) with those of α. In the substitution operation, an aligned pair of substitution 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 witho...

Ngày tải lên: 07/03/2014, 22:20

... 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 terms 4 . The statis- tics of the resulting dataset are shown in Table 1. 3.2 Semantic Categories The semantic categorie...

Ngày tải lên: 08/03/2014, 00:20

... 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 a causal classifier should label it CAUSAL since this and can be paraphrased as and as a result. We identified both s...

Ngày tải lên: 08/03/2014, 01:20

... of learning semantic categories, all the previous studies using web search logs rely on web search query logs. Therefore, we propose to use web search clickthrough 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 from search...

Ngày tải lên: 08/03/2014, 01:20

... word order in most European languages is subject-verb-object (SVO), and some words with similar forms are used with similar meanings in different languages. The reasons for these common properties ... grammatical category clustered across languages as well as within a language. Fig- ure 3 shows examples of inferred common gram- mar rules with high probabilities. Grammar rules that seem...

Ngày tải lên: 07/03/2014, 22:20

... J. DeNero, and D. Klein. 2010. Painless unsupervised learning with features. In Proc. of NAACL. P. Blunsom and T. Cohn. 2006. Discriminative word alignment 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 s a j , and the discretized jump feature conjoined with the...

Ngày tải lên: 17/03/2014, 00:20