Báo cáo khoa học: "Left-to-Right Target Generation for Hierarchical Phrase-based Translation" doc

... achieved performance of the model im- proved; the best performance achieved by Method 2 is .63 points higher than the performance of Method 1. We should note that the oracle performance for Method ... Thus we do not need to do anything further to obtain input of the form necessary for application of the inflection model. For the phrase-based system, we generated the annotations needed...

Ngày tải lên: 08/03/2014, 01:20

... 1) for j = 1 to qc(x) do for i = 1 to k do h i (x) ← hash of event {x, j} under h i BF[h i (x)] ← 1 end for end for end for return BF 3.1 Log-frequency Bloom filter The efficiency of our scheme for ... bound on qc(x) ∈ S train for j = 1 to M AXQCOUNT do for i = 1 to k do h i (x) ← hash of event {x, j} under h i if BF[h i (x)] = 0 then return j − 1 end if end for end for The...

Ngày tải lên: 17/03/2014, 04:20

... target information during the measure word generation process. We do not integrate our measure word generation module into the SMT decoder since there is only little target contextual information ... of Formula (3) involves the fea- tures listed in Table 2 where the Mo-ME model only employs target features and the Bi-ME model leverages both target features and source features...

Ngày tải lên: 08/03/2014, 01:20

... a hierarchical phrase-based SMT system as our baseline; thus, the units involved in computation of sense similarities are hierar- chical rules. 2 Hierarchical phrase-based MT system The hierarchical ... 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....

Ngày tải lên: 20/02/2014, 04:20

... error reduction. Results are for dependency parsing on the dev set for iters:5,training-k:1. tal errors break down by gold head. For example, the 12.1% total error reduction for attachments of an IN ... leave, in have NN IN time, for some NN IN money, for annual NNS IN reports, on Table 8: The highest-weight features (thresholded at a count of 400) of the before-word schema for...

Ngày tải lên: 20/02/2014, 04:20

... probabilistic model for headline generation in which they divide head- line generation process into two steps; namely the step of distilling the information source from the observation of a document and ... 0.17252 for the same set of documents. 1 Introduction A headline is considered as a condensed summary of a document. It can be classified as the acme of text summarization....

Ngày tải lên: 20/02/2014, 05:20

... Boks 1080 Blindern; 0316 Oslo (Norway) ♠ Center for the Study of Language and Information, Stanford, CA 94305 (USA) { jtl@ifi.uio.no| oe@csli.stanford.edu} Abstract The LOGON MT demonstrator assembles independently ... Norwegian Research Council program for building national infrastructure for language technology (Fenstad et al., 2006). It is the goal for the program as well as for...

Ngày tải lên: 20/02/2014, 12:20

... grouping of information, thus expressing the information in- terdependencies. The language which SUG pro- vides for specifying these descriptions allows par- tiality both in the information about ... thereby also forgetting the predications over the nodes. This forgetting operation abstracts away from the existence of the forgotten node in the phrase structure. Once a node is forgot...

Ngày tải lên: 20/02/2014, 21:20

... to improved performance for any method. One utterance of context is best for shift-reduce and start-join; three is best for the connection path method. The shift- reduce method performs the best. ... history Nbest F start!complete connection!paths shift!reduce Figure 5: Performance of parse-based methods for subtask tree building Figure 5 shows the performance of the different methods...

Ngày tải lên: 22/02/2014, 02:20

... map-free information- theoretic criterion—see Gao and Johnson (2008) for details. Although we find M-to-1 to be the most reliable criterion of the three, we include the other two criteria for completeness. ... Table 1 compares the per- formance of SVD2 to other leading models. Fol- lowing Gao and Johnson (2008), the number of induced tags is 17 for PTB17 evaluation and 50 for PTB45...

Ngày tải lên: 07/03/2014, 22:20