emnlp emnlp2012 emnlp2012-74 knowledge-graph by maker-knowledge-mining

74 emnlp-2012-Language Model Rest Costs and Space-Efficient Storage

Source: pdf

Author: Kenneth Heafield ; Philipp Koehn ; Alon Lavie

Abstract: Approximate search algorithms, such as cube pruning in syntactic machine translation, rely on the language model to estimate probabilities of sentence fragments. We contribute two changes that trade between accuracy of these estimates and memory, holding sentence-level scores constant. Common practice uses lowerorder entries in an N-gram model to score the first few words of a fragment; this violates assumptions made by common smoothing strategies, including Kneser-Ney. Instead, we use a unigram model to score the first word, a bigram for the second, etc. This improves search at the expense of memory. Conversely, we show how to save memory by collapsing probability and backoff into a single value without changing sentence-level scores, at the expense of less accurate estimates for sentence fragments. These changes can be stacked, achieving better estimates with unchanged memory usage. In order to interpret changes in search accuracy, we adjust the pop limit so that accuracy is unchanged and report the change in CPU time. In a GermanEnglish Moses system with target-side syntax, improved estimates yielded a 63% reduction in CPU time; for a Hiero-style version, the reduction is 21%. The compressed language model uses 26% less RAM while equivalent search quality takes 27% more CPU. Source code is released as part of KenLM.

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 {hea fie ld, alavie} @ c s cmu edu Abstract Approximate search algorithms, such as cube pruning in syntactic machine translation, rely on the language model to estimate probabilities of sentence fragments. [sent-3, score-0.471]

2 We contribute two changes that trade between accuracy of these estimates and memory, holding sentence-level scores constant. [sent-4, score-0.122]

3 Common practice uses lowerorder entries in an N-gram model to score the first few words of a fragment; this violates assumptions made by common smoothing strategies, including Kneser-Ney. [sent-5, score-0.263]

4 Conversely, we show how to save memory by collapsing probability and backoff into a single value without changing sentence-level scores, at the expense of less accurate estimates for sentence fragments. [sent-8, score-0.517]

5 These changes can be stacked, achieving better estimates with unchanged memory usage. [sent-9, score-0.247]

6 In order to interpret changes in search accuracy, we adjust the pop limit so that accuracy is unchanged and report the change in CPU time. [sent-10, score-0.517]

7 In a GermanEnglish Moses system with target-side syntax, improved estimates yielded a 63% reduction in CPU time; for a Hiero-style version, the reduction is 21%. [sent-11, score-0.184]

8 1 Introduction Language model storage is typically evaluated in terms of speed, space, and accuracy. [sent-14, score-0.098]

9 uk a fourth dimension, rest cost quality, that captures how well the model scores sentence fragments for purposes of approximate search. [sent-18, score-0.557]

10 Rest cost quality is distinct from accuracy in the sense that the score of a complete sentence is held constant. [sent-19, score-0.15]

11 We first show how to improve rest cost quality over standard practice by using additional space. [sent-20, score-0.326]

12 Then, conversely, we show how to compress the language model by making a pessimistic rest cost assumption1 . [sent-21, score-0.543]

13 However, approximate search algorithms use estimates for sentence fragments. [sent-23, score-0.109]

14 If the language model has order N (an N-gram model), then the first N − 1 words of the fragment have incomplete icrosntNt ext − an 1d w thoerd lsa ostf N th e− f 1ra gwmorednst h haavvee n inotbceoemnp completely ausnded t as caostn Ntex −t. [sent-24, score-0.181]

15 , 2009) that uses lower-order entries from the language model for the first words in the fragment and no rest cost adjustment for the last few words. [sent-28, score-0.653]

16 Formally, the baseline estimate for sentence fragment w1k is NnY=−11pN(wn|w1n−1)! [sent-29, score-0.222]

17 where each wn is a word and pN is an N-gram language model. [sent-31, score-0.098]

18 1Here, the term rest cost means an adjustment to the score of a sentence fragment but not to whole sentences. [sent-33, score-0.641]

19 Sentence fragments frequently begin |wn1−1 with “the”. [sent-41, score-0.101]

20 We then use pn to score the nth word of a sentence fragment. [sent-50, score-0.113]

21 Thus, a unigram model scores the first word of a sentence fragment, a bigram model scores the second word, and so on until either the n-gram is not present in the model or the first N −1 words have been scored. [sent-51, score-0.149]

22 Conversely, we can lower memory consumption relative to the baseline at the expense of poorer rest costs. [sent-53, score-0.441]

23 Baseline models store two entries per n-gram: probability and backoff. [sent-54, score-0.131]

24 We will show that the probability and backoff values in a language model can be collapsed into a single value for each n-gram without changing sentence probability. [sent-55, score-0.225]

25 This trans- formation saves memory by halving the number of values stored per entry, but it makes rest cost estimates worse. [sent-56, score-0.58]

26 Specifically, the rest cost pessimistically assumes that the model will back off to unigrams immediately following the sentence fragment. [sent-57, score-0.367]

27 To measure the impact of their different rest costs, we experiment with cube pruning (Chiang, 2007) in syntactic machine transla2Other smoothing techniques, including Witten-Bell (Witten and Bell, 1991), do not make this assumption. [sent-59, score-0.678]

28 Cube pruning’s goal is to find high-scoring sentence fragments for the root non-terminal in the parse tree. [sent-61, score-0.142]

29 It does so by going bottom-up in the parse tree, searching for high-scoring sentence fragments for each non-terminal. [sent-62, score-0.142]

30 Within each non-terminal, it generates a fixed number of high-scoring sentence fragments; this is known as the pop limit. [sent-63, score-0.379]

31 Increasing the pop limit therefore makes search more accurate but costs more time. [sent-64, score-0.769]

32 By moderating the pop limit, improved accuracy can be interpreted as a reduction in CPU time and vice-versa. [sent-65, score-0.396]

33 2 Related Work Vilar and Ney (201 1) study several modifications to cube pruning and cube growing (Huang and Chiang, 2007). [sent-66, score-0.664]

34 This first pass is cheaper because translation alternatives are likely to fall into the same class. [sent-68, score-0.111]

35 The rest cost estimates we describe here could be applied in both passes, so our work is largely orthogonal. [sent-71, score-0.394]

36 Zens and Ney (2008) present rest costs for phrasebased translation. [sent-72, score-0.516]

37 These rest costs are based on factors external to the sentence fragment, namely output that the decoder may generate in the future. [sent-73, score-0.632]

38 Our rest costs examine words internal to the sentence fragment, namely the first and last few words. [sent-74, score-0.557]

39 While data structure compression (Raj and Whittaker, 2003; Heafield, 2011) and randomized data structures (Talbot and Osborne, 2007; Guthrie and Hepple, 2010) are useful, here we are concerned solely with the values stored by these data structures. [sent-77, score-0.098]

40 Quantization (Whittaker and Raj, 2001; Federico and Bertoldi, 2006) uses less bits to store each numerical value at the expense of model quality, including scores of full sentences, and is compatible with our approach. [sent-78, score-0.271]

41 In fact, the lower-order probabilities might be quantized further than normal since these are used solely for rest cost purposes. [sent-79, score-0.416]

42 Our compression technique reduces storage from two values, probability and backoff, to one value, theoretically halving the bits per value (except N-grams which all have backoff 1). [sent-80, score-0.445]

43 This makes the storage requirement for higher-quality modified Kneser-Ney smoothing comparable to stupid backoff (Brants et al. [sent-81, score-0.408]

44 Whether to use one smoothing technique or the other then becomes largely an issue of training costs and quality after quantization. [sent-83, score-0.373]

45 1 Better Rest Costs As alluded to in the introduction, the first few words of a sentence fragment are typically scored using lower-order entries from an N-gram language model. [sent-85, score-0.275]

46 However, Kneser-Ney smoothing (Kneser and Ney, 1995) conditions lower-order probabilities on backing off. [sent-86, score-0.164]

47 This adjustment is also performed for modified Kneser-Ney smoothing. [sent-88, score-0.127]

48 In some cases, we are able to determine that the model will back off and therefore the lowerorder probability makes the appropriate assumption. [sent-94, score-0.136]

49 Specifically, if vwn1 does not appear in the model for any word v, then computing p(wn will al- |vw1n−1) 3Counts are not modified for n-grams bound to the beginning of sentence, namely those with w1 = have backoff 1, so it follows from Proposition 1that sentencelevel scores are unchanged. [sent-95, score-0.236]

50 q( w1k ) = p( w1k ) Proposition 1 characterizes q as a pessimistic rest cost on sentence fragments that scores sentences in exactly the same way as the baseline using p and b. [sent-96, score-0.685]

51 , 2008) or modify the decoder to annotate sentence fragments with backoff information (Heafield, 2011); we have effectively moved this step to preprocessing. [sent-103, score-0.365]

52 The disadvantage is that q is not a proper probability and it produces worse rest costs than does the baseline. [sent-104, score-0.516]

53 Language models are actually applied at two points in syntactic machine translation: scoring lexical items in grammar rules and during cube pruning. [sent-105, score-0.371]

54 Grammar scoring is an offline and embarrassingly parallel process where memory is not as tight (since the phrase table is streamed) and fewer queries are made, so slow non-lossy compression and even network-based sharding can be used. [sent-106, score-0.281]

55 We therefore use an ordinary language model for grammar scoring and only apply the compressed model during cube pruning. [sent-107, score-0.418]

56 Grammar scoring impacts grammar pruning (by selecting only top-scoring grammar rules) and the order in which rules are tried during cube pruning. [sent-108, score-0.649]

57 3 Combined Scheme Our two language model modifications can be trivially combined by using lower-order probabilities on the left of a fragment and by charging all backoff penalties on the right of a fragment. [sent-110, score-0.411]

58 The net result is a language model that uses the same memory as the baseline but has better rest cost estimates. [sent-111, score-0.491]

59 4 Experiments To measure the impact of different rest costs, we use the Moses chart decoder (Koehn et al. [sent-112, score-0.292]

60 Using the Moses pipeline, we trained two syntactic German-English systems, one with target-side syntax and the other hierarchical with unlabeled grammar rules (Chiang, 2007). [sent-115, score-0.197]

61 Models were built and interpolated using SRILM (Stolcke, 2002) with modified Kneser-Ney smoothing (Kneser and Ney, 1995; Chen and Goodman, 1998) and the default pruning settings. [sent-118, score-0.257]

62 For lower-order rest costs, we also built models with orders 1 through 4 then used the n-gram model to score n-grams in the 5-gram model. [sent-120, score-0.217]

63 Feature weights were trained with MERT (Och, 2003) on the baseline using a pop limit of 1000 and 100-best output. [sent-121, score-0.47]

64 1 Rest Costs as Prediction Scoring the first few words of a sentence fragment is a prediction task. [sent-125, score-0.222]

65 In order to measure performance on this task, we ran the decoder on the hierarchical system with a pop limit of 1000. [sent-127, score-0.601]

66 02a894r6 Table 1: Bias (mean error), mean squared error, and variance (of the error) for the lower-order rest cost and the baseline. [sent-136, score-0.396]

67 Statistics were computed separately for the first word of a fragment (n = 1), the second word (n = 2), etc. [sent-138, score-0.181]

68 The lowerorder estimates are better across the board, reducing error in cube pruning. [sent-139, score-0.473]

69 ity) for both lower-order rest costs and the baseline. [sent-142, score-0.516]

70 Cube pruning uses relative scores, so bias matters less, though positive bias will favor rules with more arity. [sent-144, score-0.246]

71 Variance matters the most because lower variance means cube pruning’s relative rankings are more accurate. [sent-145, score-0.306]

72 Our lower-order rest costs are better across the board in terms of absolute bias, mean squared error, and variance. [sent-146, score-0.551]

73 2 Pop Limit Trade-Offs The cube pruning pop limit is a trade-off between search accuracy and CPU time. [sent-148, score-0.857]

74 Here, we measure how our rest costs improve (or degrade) that trade-off. [sent-149, score-0.516]

75 Lower-order rest costs perform better in both systems, reaching plateau model scores and BLEU with less CPU time. [sent-158, score-0.57]

76 get syntax, where a pop limit of 50 outperforms the baseline with pop limit 700. [sent-175, score-0.94]

77 The combined setting, using the same memory as the baseline, shows a similar 62. [sent-180, score-0.132]

78 We attribute this differ- ence to improved grammar rule scoring that impacts pruning and sorting. [sent-182, score-0.329]

79 In the target syntax model, the grammar is not saturated (i. [sent-183, score-0.188]

80 less pruning will still improve scores) but we nonetheless prune for tractability reasons. [sent-185, score-0.149]

81 The lower-order rest costs are particularly useful for grammar pruning because lexical items are typically less than five words long (and frequently only word). [sent-186, score-0.747]

82 The hierarchical grammar is nearly saturated with respect to grammar pruning, so improvement there is due mostly to better search. [sent-187, score-0.267]

83 34 is achieved under the lowerorder condition with pop limits 50 and 200, while 1175 other scenarios are still climbing to the plateau. [sent-189, score-0.474]

84 With a pop limit of 1000, the baseline’s average model score is -101. [sent-190, score-0.47]

85 Better average models scores are obtained from the lower-order model with pop limit 690 using 79% of baseline CPU, the combined model with pop limit 900 using 97% CPU, and the pessimistic model with pop limit 1350 using 127% CPU. [sent-192, score-1.627]

86 Pessimistic compression does worsen search, re- quiring 27% more CPU in the hierarchical system to achieve the same quality. [sent-193, score-0.154]

87 3 Memory Usage Our rest costs add a value (for lower-order probabilities) or remove a value (pessimistic compression) for each n-gram except those of highest order (n = N). [sent-196, score-0.516]

88 and removes another, so it uses the same memory as the baseline. [sent-199, score-0.132]

89 The memory footprint of adding or removing a value depends on the number of such ngrams, the underlying data structure, and the extent of quantization. [sent-200, score-0.132]

90 Storage size ofthe smallest model is reduced by 26%, bringing higher-quality smoothed models in line with stupid backoff models that also store one value per n-gram. [sent-206, score-0.28]

91 Structure Baseline Change % Probing4,072517 13% Trie 2,647 506 19% 8-bit quantized trie 1,236 140 11% 8-bit minimal perfect hash 540 140 26% Table 2: Size in megabytes of our language model, excluding operating system overhead. [sent-208, score-0.144]

92 Change is the cost of adding an additional value to store lowerorder probabilities. [sent-209, score-0.323]

93 Efficiently storing lower-order probabilities and using them as rest costs improves both cube pruning (21% CPU reduction in a hierarchical system) and model filtering (net 63% CPU time reduction with target syntax) at the expense of 13-26% more RAM for the language model. [sent-212, score-1.252]

94 This model filtering improvement is surprising both in the impact relative to changing the pop limit and simplicity of implementation, since it can be done offline. [sent-213, score-0.506]

95 Compressing the language model to halve the number of values per n-gram (except Ngrams) results in a 13-26% reduction in RAM with 26% over the smallest model, costing 27% more CPU and leaving overall sentence scores unchanged. [sent-214, score-0.153]

96 This compression technique is likely to have more general application outside of machine translation, especially where only sentence-level scores are required. [sent-215, score-0.152]

97 How many bits are needed to store probabilities for phrasebased translation? [sent-257, score-0.168]

98 A scalable decoder for parsing-based machine translation with equivalent language model state maintenance. [sent-294, score-0.149]

99 Lossless compression of language model structure and word identifiers. [sent-312, score-0.098]

100 Cardinality pruning and language model heuristics for hierarchical phrase-based translation. [sent-324, score-0.205]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('pop', 0.338), ('cpu', 0.313), ('costs', 0.299), ('cube', 0.238), ('rest', 0.217), ('fragment', 0.181), ('pessimistic', 0.163), ('pruning', 0.149), ('backoff', 0.148), ('lowerorder', 0.136), ('memory', 0.132), ('limit', 0.132), ('heafield', 0.117), ('cost', 0.109), ('fragments', 0.101), ('wn', 0.098), ('compression', 0.098), ('storage', 0.098), ('kneser', 0.093), ('adjustment', 0.093), ('expense', 0.092), ('ram', 0.084), ('grammar', 0.082), ('guthrie', 0.081), ('whittaker', 0.081), ('store', 0.078), ('decoder', 0.075), ('smoothing', 0.074), ('translation', 0.074), ('federico', 0.072), ('pn', 0.072), ('raj', 0.07), ('estimates', 0.068), ('bleu', 0.066), ('kenlm', 0.063), ('moses', 0.063), ('koehn', 0.063), ('marcello', 0.06), ('syntax', 0.059), ('reduction', 0.058), ('philipp', 0.056), ('hierarchical', 0.056), ('hash', 0.055), ('belazzougui', 0.054), ('bhiksha', 0.054), ('compress', 0.054), ('displace', 0.054), ('halving', 0.054), ('stupid', 0.054), ('trestles', 0.054), ('zhifei', 0.054), ('scores', 0.054), ('entries', 0.053), ('scoring', 0.051), ('chris', 0.051), ('dyer', 0.049), ('ney', 0.047), ('unchanged', 0.047), ('bits', 0.047), ('quantized', 0.047), ('saturated', 0.047), ('backing', 0.047), ('compressed', 0.047), ('impacts', 0.047), ('nicola', 0.045), ('wmt', 0.045), ('europarl', 0.043), ('conversely', 0.043), ('probabilities', 0.043), ('weese', 0.042), ('trie', 0.042), ('omar', 0.042), ('vilar', 0.042), ('sanjeev', 0.042), ('hepple', 0.042), ('bertoldi', 0.042), ('zens', 0.042), ('storing', 0.042), ('sentence', 0.041), ('hermann', 0.04), ('modifications', 0.039), ('kenneth', 0.039), ('proposition', 0.039), ('chiang', 0.037), ('pass', 0.037), ('acoustics', 0.037), ('srilm', 0.037), ('changing', 0.036), ('uk', 0.035), ('variance', 0.035), ('squared', 0.035), ('hashing', 0.035), ('modified', 0.034), ('edinburgh', 0.034), ('net', 0.033), ('joshua', 0.033), ('matters', 0.033), ('bias', 0.032), ('toolkit', 0.032), ('error', 0.031)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0000001 74 emnlp-2012-Language Model Rest Costs and Space-Efficient Storage

Author: Kenneth Heafield ; Philipp Koehn ; Alon Lavie

2 0.20537308 11 emnlp-2012-A Systematic Comparison of Phrase Table Pruning Techniques

Author: Richard Zens ; Daisy Stanton ; Peng Xu

Abstract: When trained on very large parallel corpora, the phrase table component of a machine translation system grows to consume vast computational resources. In this paper, we introduce a novel pruning criterion that places phrase table pruning on a sound theoretical foundation. Systematic experiments on four language pairs under various data conditions show that our principled approach is superior to existing ad hoc pruning methods.

3 0.1562833 57 emnlp-2012-Generalized Higher-Order Dependency Parsing with Cube Pruning

Author: Hao Zhang ; Ryan McDonald

Abstract: State-of-the-art graph-based parsers use features over higher-order dependencies that rely on decoding algorithms that are slow and difficult to generalize. On the other hand, transition-based dependency parsers can easily utilize such features without increasing the linear complexity of the shift-reduce system beyond a constant. In this paper, we attempt to address this imbalance for graph-based parsing by generalizing the Eisner (1996) algorithm to handle arbitrary features over higherorder dependencies. The generalization is at the cost of asymptotic efficiency. To account for this, cube pruning for decoding is utilized (Chiang, 2007). For the first time, label tuple and structural features such as valencies can be scored efficiently with third-order features in a graph-based parser. Our parser achieves the state-of-art unlabeled accuracy of 93.06% and labeled accuracy of 91.86% on the standard test set for English, at a faster speed than a reimplementation ofthe third-ordermodel of Koo et al. (2010).

4 0.14575012 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

Abstract: Independence between sentences is an assumption deeply entrenched in the models and algorithms used for statistical machine translation (SMT), particularly in the popular dynamic programming beam search decoding algorithm. This restriction is an obstacle to research on more sophisticated discourse-level models for SMT. We propose a stochastic local search decoding method for phrase-based SMT, which permits free document-wide dependencies in the models. We explore the stability and the search parameters ofthis method and demonstrate that it can be successfully used to optimise a document-level semantic language model. 1 Motivation In the field oftranslation studies, it is undisputed that discourse-wide context must be considered care- fully for good translation results (Hatim and Mason, 1990). By contrast, the state of the art in statistical machine translation (SMT), despite significant advances in the last twenty years, still assumes that texts can be translated sentence by sentence under strict independence assumptions, even though it is well known that certain linguistic phenomena such as pronominal anaphora cannot be translated correctly without referring to extra-sentential context. This is true both for the phrase-based and the syntaxbased approach to SMT. In the rest of this paper, we shall concentrate on phrase-based SMT. One reason why it is difficult to experiment with document-wide models for phrase-based SMT is that the dynamic programming (DP) algorithm 1179 which has been used almost exclusively for decoding SMT models in the recent literature has very strong assumptions of locality built into it. DP beam search for phrase-based SMT was described by Koehn et al. (2003), extending earlier work on word-based SMT (Tillmann et al., 1997; Och et al., 2001 ; Tillmann and Ney, 2003). This algorithm con- structs output sentences by starting with an empty hypothesis and adding output words at the end until translations for all source words have been generated. The core models of phrase-based SMT, in particular the n-gram language model (LM), only depend on a constant number of output words to the left of the word being generated. This fact is exploited by the search algorithm with a DP technique called hypothesis recombination (Och et al., 2001), which permits the elimination of hypotheses from the search space if they coincide in a certain number of final words with a better hypothesis and no future expansion can possibly invert the relative ranking of the two hypotheses under the given models. Hypothesis recombination achieves a substantial reduction of the search space without affecting search optimality and makes it possible to use aggressive pruning techniques for fast search while still obtaining good results. The downside of this otherwise excellent approach is that it only works well with models that have a local dependency structure similar to that of an n-gram language model, so they only depend on a small context window for each target word. Sentence-local models with longer dependencies can be added, but doing so greatly increases the risk for search errors by inhibiting hypothesis recombination. Cross-sentence dependencies cannot be directly integrated into DP SMT decoding in LParnogcue agdein Lgesa ornf tihneg, 2 p0a1g2e Jso 1in17t C9–o1n1f9e0re,n Jce ju on Is Elanmdp,ir Kicoarlea M,e 1t2h–o1d4s J iunly N 2a0tu1r2a.l ? Lc a2n0g1u2ag Aes Psorcoicaetsiosin fgo arn Cdo Cmopmutpauti oantiaoln Lailn Ngautiustriacls any obvious way, especially if joint optimisation of a number of interdependent decisions over an entire document is required. Research into models with a more varied, non-local dependency structure is to some extent stifled by the difficulty of decoding such models effectively, as can be seen by the problems some researchers encountered when they attempted to solve discourse-level problems. Consider, for instance, the work on cache-based language models by Tiedemann (2010) and Gong et al. (201 1), where error propagation was a serious issue, or the works on pronominal anaphora by Le Nagard and Koehn (2010), who implemented cross-sentence dependencies with an ad-hoc two-pass decoding strategy, and Hardmeier and Federico (2010) with the use of an external decoder driver to manage backward-only dependencies between sentences. In this paper, we present a method for decoding complete documents in phrase-based SMT. Our decoder uses a local search approach whose state consists of a complete translation of an entire document at any time. The initial state is improved by the application of a series of operations using a hill climbing strategy to find a (local) maximum of the score function. This setup gives us complete freedom to define scoring functions over the entire document. Moreover, by optionally initialising the state with the output of a traditional DP decoder, we can ensure that the final hypothesis is no worse than what would have been found by DP search alone. We start by describing the decoding algorithm and the state operations used by our decoder, then we present empirical results demonstrating the effectiveness of our approach and its usability with a document-level semantic language model, and finally we discuss some related work. 2 SMT Decoding by Hill Climbing In this section, we formally describe the phrasebased SMT model implemented by our decoder as well as the decoding algorithm we use. 2.1 SMT Model Our decoder is based on local search, so its state at any time is a representation of a complete translation of the entire document. Even though the decoder operates at the document level, it is important to keep 1180 track of sentence boundaries, and the individual operations that are applied to the state are still confined to sentence scope, so it is useful to decompose the state of a document into the state of its sentences, and we define the overall state S as a sequence of sentence states: S = S1S2 . . .SN, (1) where N is the number of sentences. This implies that we constrain the decoder to emit exactly one output sentence per input sentence. Let ibe the number of a sentence and mi the number of input tokens of this sentence, p and q (with 1 ≤ p ≤ q ≤ mi) be positions in the input sentence a1n ≤d [p; q] qde ≤no mte the set ofpositions from p up to and including q. We say that [p; q] precedes [p0; q0], or [p; q] ≺ [p0; q0], if q < p0. Let Φi([p; q]) be the set of t[pra;nqs]l ≺atio [pns for the source phrase covering positions [p; q] in the input sentence ias given by the phrase table. We call A = h[p; q] ,φi an anchored phrase pair w.it Wh coverage C(A) = [p; q] nif a φ ∈ Φi([p; q]) sise a target phrase translating =th [ep source w∈o Φrds at positions [p; q] . Then a sequence of ni anchored phrase pairs Si = A1A2 . . .Ani (2) is a valid sentence state for sentence iif the following two conditions hold: 1. The coverage sets C(Aj) for j in 1, . . . , ni are mutually disjoint, and 2. the anchored phrase pairs jointly cover the complete input sentence, or [niC(Aj) = [1;mi]. (3) [j=1 Let f(S) be a scoring function mapping a state S to a real number. As usual in SMT, it is assumed that the scoring function can be decomposed into a linear combination of K feature functions hk(S), each with a constant weight λk, so f(S) =k∑K=1λkhk(S). (4) The problem addressed by the decoder is the search for the state with maximal score, such that Sˆ Sˆ = argSmaxf(S). (5) The feature functions implemented in our baseline system are identical to the ones found in the popular Moses SMT system (Koehn et al., 2007). In particular, our decoder has the following feature functions: 1. phrase translation scores provided by the phrase table including forward and backward conditional probabilities, lexical weights and a phrase penalty (Koehn et al., 2003), 2. n-gram language model scores implemented with the KenLM toolkit (Heafield, 2011), 3. a word penalty score, 4. a distortion model with geometric (Koehn et al., 2003), and decay 5. a feature indicating the number of times a given distortion limit is exceeded in the current state. In our experiments, the last feature is used with a fixed weight of negative infinity in order to limit the gaps between the coverage sets of adjacent anchored phrase pairs to a maximum value. In DP search, the distortion limit is usually enforced directly by the search algorithm and is not added as a feature. In our decoder, however, this restriction is not required to limit complexity, so we decided to add it among the scoring models. 2.2 Decoding Algorithm The decoding algorithm we use (algorithm 1) is very simple. It starts with a given initial document state. In the main loop, which extends from line 3 to line 12, it generates a successor state S0 for the current state S by calling the function Neighbour, which non-deterministically applies one of the operations described in section 3 of this paper to S. The score of the new state is compared to that of the previous one. If it meets a given acceptance criterion, S0 becomes the current state, else search continues from the previous state S. For the experiments in this paper, we use the hill climbing acceptance criterion, which simply accepts a new state if its score is higher than that of the current state. Other acceptance criteria are possible and could be used to endow the search algorithm with stochastic behaviour. 1181 The main loop is repeated until a maximum number of steps (step limit) is reached or until a maximum number of moves are rejected in a row (rejection limit). Algorithm 1 Decoding algorithm Input: an initial document state S; search parameters maxsteps and maxrejected Output: a modified document state 1: nsteps ← 0 2: nrejected ← 0 3: nwrhejileec nsteps < maxsteps and nrejected < maxrejected do 4: S0 ← Neighbour (S) 5: if Accept (f(S0) , f(S)) then 6: S ← S0 7: nrejected ← 0 8: elsner 9: nrejected ← nrejected + 1 10: enndr eifj 11: nsteps ← nsteps + 1 12: ennds wtephsile ← 13: return S A notable difference between this algorithm and other hill climbing algorithms that have been used for SMT decoding (Germann et al., 2004; Langlais et al., 2007) is its non-determinism. Previous work for sentence-level decoding employed a steepest ascent strategy which amounts to enumerating the complete neighbourhood of the current state as defined by the state operations and selecting the next state to be the best state found in the neighbourhood of the current one. Enumerating all neighbours of a given state, costly as it is, has the advantage that it makes it easy to prove local optimality of a state by recognising that all possible successor states have lower scores. It can be rather inefficient, since at every step only one modification will be adopted; many of the modifications that are discarded will very likely be generated anew in the next iteration. As we extend the decoder to the document level, the size of the neighbourhood that would have to be explored in this way increases considerably. Moreover, the inefficiency of the steepest ascent approach potentially increases as well. Very likely, a promising move in one sentence will remain promising after a modification has been applied to another sentence, even though this is not guaranteed to be true in the presence of cross-sentence models. We therefore adopt a first-choice hill climbing strategy that non-deterministically generates successor states and accepts the first one that meets the acceptance criterion. This frees us from the necessity of generating the full set of successors for each state. On the downside, if the full successor set is not known, it is no longer possible to prove local optimality of a state, so we are forced to use a different condition for halting the search. We use a combination of two limits: The step limit is a hard limit on the resources the user is willing to expend on the search problem. The value of the rejection limit determines how much of the neighbourhood is searched for better successors before a state is accepted as a solution; it is related to the probability that a state returned as a solution is in fact locally optimal. To simplify notations in the description of the individual state operations, we write Si −→ Si0 (6) to signify that a state operation, when presented with a document state as in equation 1 and acting on sentence i, returns a new document state of S0 = S1 . . .Si−1 Si0 Si+1 . . .SN. (7) Similarly, Si : Aj . . .Aj+h−1 −→ A01 . . .A0h0 (8) is equivalent to Si −→ A1 . . .Aj−1 A01 . . .A0h0 Aj+h . . .Ani (9) and indicates that the operation returns a state in which a sequence of h consecutive anchored phrase pairs has been replaced by another sequence of h0 anchored phrase pairs. 2.3 Efficiency Considerations When implementing the feature functions for the decoder, we have to exercise some care to avoid recomputing scores for the whole document at every iteration. To achieve this, the scores are computed completely only once, at the beginning of the decoding run. In subsequent iterations, scoring functions are presented with the scores of the previous 1182 iteration and a list of modifications produced by the state operation, a set of tuples hi, r, s,A01 . . .A0h0i, each indicating tthioant ,t ahe s edto ocfu tmupelnets s hhio,ru,sld, Abe modifii,e eda as described by Si :Ar . . .As −→ A01 . . .A0h0 . (10) If a feature function is decomposable in some way, as all the standard features developed under the constraints of DP search are, it can then update the state simply by subtracting and adding score components pertaining to the modified parts of the document. Feature functions have the possibility to store their own state information along with the document state to make sure the required information is available. Thus, the framework makes it possible to exploit decomposability for efficient scoring without impos- ing any particular decomposition on the features as beam search does. To make scoring even more efficient, scores are computed in two passes: First, every feature function is asked to provide an upper bound on the score that will be obtained for the new state. In some cases, it is possible to calculate reasonable upper bounds much more efficiently than computing the exact feature value. If the upper bound fails to meet the acceptance criterion, the new state is discarded right away; if not, the full score is computed and the acceptance criterion is tested again. Among the basic SMT models, this two-pass strategy is only used for the n-gram LM, which requires fairly expensive parameter lookups for scoring. The scores of all the other baseline models are fully computed during the first scoring pass. The n-gram model is more complex. In its state information, it keeps track of the LM score and LM library state for each word. The first scoring pass then identifies the words whose LM scores are affected by the current search step. This includes the words changed by the search operation as well as the words whose LM history is modified. The range of the history de- pendencies can be determined precisely by considering the “valid state length” information provided by the KenLM library. In the first pass, the LM scores of the affected words are subtracted from the total score. The model only looks up the new LM scores for the affected words and updates the total score if the new search state passes the first acceptance check. This two-pass scoring approach allows us to avoid LM lookups altogether for states that will be rejected anyhow because of low scores from the other models, e. g. because the distortion limit is violated. Model score updates become more complex and slower as the number of dependencies of a model increases. While our decoding algorithm does not impose any formal restrictions on the number or type of dependencies that can be handled, there will be practical limits beyond which decoding becomes unacceptably slow or the scoring code becomes very difficult to maintain. These limits are however fairly independent of the types of dependencies handled by a model, which permits the exploration of more varied model types than those handled by DP search. 2.4 State Initialisation Before the hill climbing decoding algorithm can be run, an initial state must be generated. The closer the initial state is to an optimum, the less work remains to be done for the algorithm. If the algorithm is to be self-contained, initialisation must be relatively uninformed and can only rely on some general prior assumptions about what might be a good initial guess. On the other hand, if optimal results are sought after, it pays off to invest some effort into a good starting point. One way to do this is to run DP search first. For uninformed initialisation, we chose to implement a very simple procedure based only on the observation that, at least for language pairs involving the major European languages, it is usually a good guess to keep the word order of the output very similar to that of the input. We therefore create the initial state by selecting, for each sentence in the document, a sequence of anchored phrase pairs covering the input sentence in monotonic order, that is, such that for all pairs of adjacent anchored phrase pairs Aj and Aj+1, we have that C(Aj) ≺ C(Aj+1 ). For initialisation with DP search, we first run the Moses decoder (Koehn et al., 2007) with default search parameters and the same models as those used by our decoder. Then we extract the best output hypothesis from the search graph of the decoder and map it into a sequence of anchored phrase pairs in the obvious way. When the document-level decoder is used with models that are incompatible with beam search, Moses can be run with a subset of the models in order to find an approximation of the solution 1183 which is then refined with the complete feature set. 3 State Operations Given a document state S, the decoder uses a neighbourhood function Neighbour to simulate a move in the state space. The neighbourhood function nondeterministically selects a type of state operation and a location in the document to apply it to and returns the resulting new state. We use a set of three operations that has the property that every possible document state can be reached from every other state in a sequence of moves. Designing operations for state transitions in local search for phrase-based SMT is a problem that has been addressed in the literature (Langlais et al., 2007; Arun et al., 2010). Our decoder’s first- choice hill climbing strategy never enumerates the full neighbourhood of a state. We therefore place less emphasis than previous work on defining a compact neighbourhood, but allow the decoder to make quite extensive changes to a state in a single step with a certain probability. Otherwise our operations are similar to those used by Arun et al. (2010). All of the operations described in this paper make changes to a single sentence only. Each time it is called, the Neighbour function selects a sentence in the document with a probability proportional to the number of input tokens in each sentence to ensure a fair distribution ofthe decoder’s attention over the words in the document regardless of varying sentence lengths. 3.1 Changing Phrase Translations The change-phrase-translation operation replaces the translation of a single phrase with a random translation with the same coverage taken from the phrase table. Formally, the operation selects an anchored phrase pair Aj by drawing uniformly from the elements of Si and then draws a new translation φ0 uniformly from the set Φi(C(Aj)). The new state is given by Si : Aj −→ hC(Aj), φ0i. (11) 3.2 Changing Word Order The swap-phrases operation affects the output word order without changing the phrase translations. It exchanges two anchored phrase pairs Aj and Aj+h, resulting in an output state of Si : Aj . . .Aj+h −→ Aj+h Aj+1 . . .Aj+h−1 Aj. (12) The start location j is drawn uniformly from the eligible sentence positions; the swap range h comes from a geometric distribution with configurable decay. Other word-order changes such as a one-way move operation that does not require another movement in exchange or more advanced permutations can easily be defined. 3.3 Resegmentation The most complex operation is resegment, which allows the decoder to modify the segmentation ofthe source phrase. It takes a number of anchored phrase pairs that form a contiguous block both in the input and in the output and replaces them with a new set of phrase pairs covering the same span of the input sentence. Formally, Si : Aj . . .Aj+h−1 −→ A01 . . .A0h0 (13) such that j+[h−1 [h0 [ C(Aj0) = [ C(A0j0) = [p;q] j[0=j (14) j[0=1 for some p and q, where, for j0 = 1, . . . ,h0, we have that A0j0 = h[pj0; qj0] , φj0i, all [pj0; qj0] are mutually disjoint =an hd[ peach φj0 isi randomly drawn from Φi([pj0;qj0]). Regardless of the ordering of Aj . . .Aj+h−1 , the resegment operation always generates a sequence of anchored phrase pairs in linear order, such that C(A0j0) ≺ C(A0j0+1 ) for j0 = 1, . . . ,h0 −1 . As )f o≺r Cth(eA other operations, j is− generated uniformly and h is drawn from a geometric distribution with a decay parameter. The new segmentation is generated by extending the sequence of anchored phrase pairs with random elements starting at the next free position, proceeding from left to right until the whole range [p; q] is covered. 4 Experimental Results In this section, we present the results of a series of experiments with our document decoder. The 1184 goal of our experiments is to demonstrate the behaviour of the decoder and characterise its response to changes in the fundamental search parameters. The SMT models for our experiments were created with a subset of the training data for the English-French shared task at the WMT 2011workshop (Callison-Burch et al., 2011). The phrase table was trained on Europarl, news-commentary and UN data. To reduce the training data to a manageable size, singleton phrase pairs were removed before the phrase scoring step. Significance-based filtering (Johnson et al., 2007) was applied to the resulting phrase table. The language model was a 5gram model with Kneser-Ney smoothing trained on the monolingual News corpus with IRSTLM (Federico et al., 2008). Feature weights were trained with Minimum Error-Rate Training (MERT) (Och, 2003) on the news-test2008 development set using the DP beam search decoder and the MERT implementation of the Moses toolkit (Koehn et al., 2007). Experimental results are reported for the newstest2009 test set, a corpus of 111 newswire documents totalling 2,525 sentences or 65,595 English input tokens. 4.1 Stability An important difference between our decoder and the classical DP decoder as well as previous work in SMT decoding with local search is that our decoder is inherently non-deterministic. This implies that repeated runs of the decoder with the same search parameters, input and models will not, in general, find the same local maximum of the score space. The first empirical question we ask is therefore how different the results are under repeated runs. The results in this and the next section were obtained with random state initialisation, i. e. without running the DP beam search decoder. Figure 1 shows the results of 7 decoder runs with the models described above, translating the newstest2009 test set, with a step limit of 227 and a rejection limit of 100,000. The x-axis of both plots shows the number of decoding steps on a logarithmic scale, so the number of steps is doubled between two adjacent points on the same curve. In the left plot, the y-axis indicates the model score optimised by the decoder summed over all 2525 sentences of the document. In the right plot, the case-sensitive BLEU score (Papineni et al., 2002) of the current decoder Figure 1: Score stability in repeated decoder runs state against a reference translation is displayed. We note, as expected, that the decoder achieves a considerable improvement of the initial state with diminishing returns as decoding continues. Between 28 and 214 steps, the score increases at a roughly logarithmic pace, then the curve flattens out, which is partly due to the fact that decoding for some documents effectively stopped when the maximum number of rejections was reached. The BLEU score curve shows a similar increase, from an initial score below 5 % to a maximum of around 21.5 %. This is below the score of 22.45 % achieved by the beam search decoder with the same models, which is not surprising considering that our decoder approximates a more difficult search problem, from which a number of strong independence assumptions have been lifted, without, at the moment, having any stronger models at its disposal to exploit this additional freedom for better translation. In terms of stability, there are no dramatic differences between the decoder runs. Indeed, the small differences that exist are hardly discernible in the plots. The model scores at the end of the decoding run range between −158767.9 and −158716.9, a g re rlautniv rea ndgieffe breetnwceee nof − only a6b7.o9ut a n0d.0 −3 %15.8 F1i6n.a9l, BLEU scores range from 21.41 % to 21.63 %, an interval that is not negligible, but comparable to the variance observed when, e. g., feature weights from repeated MERT runs are used with one and the same SMT system. Note that these results were obtained with random state initialisation. With DP initialisation, score differences between repeated runs rarely 1185 exceed 0.02 absolute BLEU percentage points. Overall, we conclude that the decoding results of our algorithm are reasonably stable despite the nondeterminism inherent in the procedure. In our subsequent experiments, the evaluation scores reported are calculated as the mean of three runs for each experiment. 4.2 Search Algorithm Parameters The hill climbing algorithm we use has two parameters which govern the trade-off between decoding time and the accuracy with which a local maximum is identified: The step limit stops the search process after a certain number of steps regardless of the search progress made or lack thereof. The rejection limit stops the search after a certain number of unsuccessful attempts to make a step, when continued search does not seem to be promising. In most of our experiments, we used a step limit of 227 ≈ 1.3 · 108 and a rejection limit of 105. In practice, decoding terminates by reaching the rejection limit for the vast majority of documents. We therefore examined the effect of different rejection limits on the learning curves. The results are shown in figure 2. The results show that continued search does pay off to a certain extent. Indeed, the curve for rejection limit 107 seems to indicate that the model score increases roughly logarithmically, albeit to a higher base, even after the curve has started to flatten out at 214 steps. At a certain point, however, the probability of finding a good successor state drops rather sharply by about two orders of magnitude, as Figure 2: Search performance at different rejection limits evidenced by the fact that a rejection limit of 106 does not give a large improvement over one of 105, while one of 107 does. The continued model score improvement also results in an increase in BLEU scores, and with a BLEU score of 22. 1% the system with rejection limit 107 is fairly close to the score of 22.45 % obtained by DP beam search. Obviously, more exact search comes at a cost, and in this case, it comes at a considerable cost, which is an explosion of the time required to decode the test set from 4 minutes at rejection limit 103 to 224 minutes at rejection limit 105 and 38 hours 45 minutes at limit 107. The DP decoder takes 3 1 minutes for the same task. We conclude that the rejection limit of 105 selected for our experiments, while technically suboptimal, realises a good trade-off between decoding time and accuracy. 4.3 A Semantic Document Language Model In this section, we present the results of the application of our decoder to an actual SMT model with cross-sentence features. Our model addresses the problem of lexical cohesion. In particular, it rewards the use of semantically related words in the translation output by the decoder, where semantic distance is measured with a word space model based on Latent Semantic Analysis (LSA). LSA has been applied to semantic language modelling in previous research with some success (Coccaro and Jurafsky, 1998; Bellegarda, 2000; Wandmacher and Antoine, 2007). In SMT, it has mostly been used for domain adaptation (Kim and Khudanpur, 2004; Tam et al., 1186 2007), or to measure sentence similarities (Banchs and Costa-juss a`, 2011). The model we use is inspired by Bellegarda (2000). It is a Markov model, similar to a standard n-gram model, and assigns to each content word a score given a history of n preceding content words, where n = 30 below. Scoring relies on a 30dimensional LSA word vector space trained with the S-Space software (Jurgens and Stevens, 2010). The score is defined based on the cosine similarity between the word vector of the predicted word and the mean word vector of the words in the history, which is converted to a probability by histogram lookup as suggested by Bellegarda (2000). The model is structurally different from a regular n-gram model in that word vector n-grams are defined over content words occurring in the word vector model only and can cross sentence boundaries. Stop words, identified by an extensive stop word list and amounting to around 60 % of the tokens, are scored by a different mechanism based on their relative frequency (undiscounted unigram probability) in the training corpus. In sum, the score produced by the semantic document LM has the following form: wh(er|h)α=is tεpαheuncipgors(wp)o|hrtinof w fci os nakutneskotn wpon w ,onerldse,in(ls1teh5) training corpus and ε is a small fixed probability. It is integrated into the decoder as an extra feature function. Since we lack an automatic method for training the feature weights of document-wide features, its weight was selected by grid search over a number of values, comparing translation performance for the newstest2009 test set. In these experiments, we used DP beam search to initialise the state of our local search decoder. Three results are presented (table 1): The first table row shows the baseline performance using DP beam search with standard sentence-local features only. The scores in the second row were obtained by running the hill climbing decoder with DP initialisation, but without adding any models. A marginal increase in scores for all three test sets demonstrates that the hill climbing decoder manages to fix some of the search errors made by the DP search. The last row contains the scores obtained by adding in the semantic language model. Scores are presented for three publicly available test sets from recent WMT Machine Translation shared tasks, of which one (newstest2009) was used to monitor progress during development and select the final model. Adding the semantic language model results in a small increase in NIST scores (Doddington, 2002) for all three test sets as well as a small BLEU score gain (Papineni et al., 2002) for two out of three corpora. We note that the NIST score turned out to react more sensitively to improvements due to the semantic LM in all our experiments, which is reasonable because the model specifically targets content words, which benefit from the information weighting done by the NIST score. While the results we present do not constitute compelling evidence in favour of our semantic LM in its current form, they do suggest that this model could be improved to realise higher gains from cross-sentence semantic information. They support our claim that cross- sentence models should be examined more closely and that existing methods should be adapted to deal with them, a problem addressed by our main contribution, the local search document decoder. 5 Related Work Even though DP beam search (Koehn et al., 2003) has been the dominant approach to SMT decoding in recent years, methods based on local search have been explored at various times. For word-based SMT, greedy hill-climbing techniques were advo1187 cated as a faster replacement for beam search (Germann et al., 2001 ; Germann, 2003; Germann et al., 2004), and a problem formulation specifically targeting word reordering with an efficient word reordering algorithm has been proposed (Eisner and Tromble, 2006). A local search decoder has been advanced as a faster alternative to beam search also for phrasebased SMT (Langlais et al., 2007; Langlais et al., 2008). That work anticipates many of the features found in our decoder, including the use of local search to refine an initial hypothesis produced by DP beam search. The possibility of using models that do not fit well into the beam search paradigm is mentioned and illustrated with the example of a reversed n-gram language model, which the authors claim would be difficult to implement in a beam search decoder. Similarly to the work by Germann et al. (2001), their decoder is deterministic and explores the entire neighbourhood of a state in order to identify the most promising step. Our main contribution with respect to the work by Langlais et al. (2007) is the introduction of the possibility of handling document-level models by lifting the assumption of sentence independence. As a consequence, enumerating the entire neighbourhood becomes too expensive, which is why we resort to a “first-choice” strategy that non-deterministically generates states and accepts the first one encountered that meets the acceptance criterion. More recently, Gibbs sampling was proposed as a way to generate samples from the posterior distribution of a phrase-based SMT decoder (Arun et al., 2009; Arun et al., 2010), a process that resembles local search in its use of a set of state-modifying operators to generate a sequence of decoder states. Where local search seeks for the best state attainable from a given initial state, Gibbs sampling produces a representative sample from the posterior. Like all work on SMT decoding that we know of, the Gibbs sampler presented by Arun et al. (2010) assumes independence of sentences and considers the complete neighbourhood of each state before taking a sample. 6 Conclusion In the last twenty years of SMT research, there has been a strong assumption that sentences in a text newstest2009 newstest2010 newstest201 1 BLEU NIST BLEU NIST BLEU NIST 22.56 6.513 27.27 7.034 24.94 7.170 + hill climbing 22.60 6.518 27.33 7.046 24.97 7.169 with semantic LM 22.71 6.549 27.53 7.087 24.90 7.199 DP search only DP Table 1: Experimental results with a cross-sentence semantic language model are independent of one another, and discourse context has been largely neglected. Several factors have contributed to this. Developing good discourse-level models is difficult, and considering the modest translation quality that has long been achieved by SMT, there have been more pressing problems to solve and lower hanging fruit to pick. However, we argue that the popular DP beam search algorithm, which delivers excellent decoding performance, but imposes a particular kind of local dependency structure on the feature models, has also had its share in driving researchers away from discourse-level problems. In this paper, we have presented a decoding procedure for phrase-based SMT that makes it possible to define feature models with cross-sentence dependencies. Our algorithm can be combined with DP beam search to leverage the quality of the traditional approach with increased flexibility for models at the discourse level. We have presented preliminary results on a cross-sentence semantic language model addressing the problem of lexical cohesion to demonstrate that this kind of models is worth exploring further. Besides lexical cohesion, cross-sentence models are relevant for other linguistic phenomena such as pronominal anaphora or verb tense selection. We believe that SMT research has reached a point of maturity where discourse phenomena should not be ignored any longer, and we consider our decoder to be a step towards this goal. References Abhishek Arun, Chris Dyer, Barry Haddow, Phil Blunsom, Adam Lopez, and Philipp Koehn. 2009. Monte carlo inference and maximization for phrase-based translation. In Proceedings of the Thirteenth Conference on Computational Natural Language Learning (CoNLL-2009), pages 102–1 10, Boulder, Colorado, June. Association for Computational Linguistics. Abhishek Arun, Barry Haddow, Philipp Koehn, Adam Lopez, Chris Dyer, and Phil Blunsom. 2010. Monte 1188 Ma- Carlo techniques for phrase-based translation. chine translation, 24(2): 103–121 . Rafael E. Banchs and Marta R. Costa-juss a`. 2011. A semantic feature for Statistical Machine Translation. In Proceedings of Fifth Workshop on Syntax, Semantics and Structure in Statistical Translation, pages 126– 134, Portland, Oregon, USA, June. Association for Computational Linguistics. Jerome R. Bellegarda. 2000. Exploiting latent semantic information in statistical language modeling. Proceedings of the IEEE, 88(8): 1279–1296. Chris Callison-Burch, Philipp Koehn, Christof Monz, and Omar Zaidan. 2011. Findings of the 2011 Workshop on Statistical Machine Translation. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 22–64, Edinburgh, Scotland, July. Association for Computational Linguistics. Noah Coccaro and Daniel Jurafsky. 1998. Towards better integration of semantic predictors in statistical language modeling. In Proceedings of the 5th International Conference on Spoken Language Processing, Sydney. George Doddington. 2002. Automatic evaluation of machine translation quality using n-gram co-occurrence statistics. In Proceedings of the second International conference on Human Language Technology Research, pages 138–145, San Diego. Jason Eisner and Roy W. Tromble. 2006. Local search with very large-scale neighborhoods for optimal permutations in machine translation. In Proceedings of the HLT-NAACL Workshop on Computationally Hard Problems and Joint Inference in Speech and Language Processing, pages 57–75. Marcello Federico, Nicola Bertoldi, and Mauro Cettolo. 2008. IRSTLM: an open source toolkit for handling large scale language models. In Interspeech 2008, pages 1618–1621 . ISCA. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2001 . Fast decoding and optimal decoding for machine translation. In Proceedings of 39th Annual Meeting of the Association for Computational Linguistics, pages 228–235, Toulouse, France, July. Association for Computational Linguis- tics. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2004. Fast and optimal decoding for machine translation. Artificial Intelligence, 154(1–2): 127–143. Ulrich Germann. 2003. Greedy decoding for Statistical Machine Translation in almost linear time. In Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics. Zhengxian Gong, Min Zhang, and Guodong Zhou. 2011. Cache-based document-level Statistical Machine Translation. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 909–919, Edinburgh, Scotland, UK., July. Association for Computational Linguistics. Christian Hardmeier and Marcello Federico. 2010. Modelling Pronominal Anaphora in Statistical Machine Translation. In Proceedings of the seventh International Workshop on Spoken Language Translation (IWSLT), pages 283–289. Basil Hatim and Ian Mason. 1990. Discourse and the Translator. Language in Social Life Series. Longman, London. Kenneth Heafield. 2011. KenLM: faster and smaller language model queries. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 187–197, Edinburgh, Scotland, July. Association for Computational Linguistics. Howard Johnson, Joel Martin, George Foster, and Roland Kuhn. 2007. Improving translation quality by discarding most of the phrasetable. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL), pages 967– 975, Prague, Czech Republic, June. Association for Computational Linguistics. David Jurgens and Keith Stevens. 2010. The S-Space package: An open source package for word space models. In Proceedings of the ACL 2010 System Demonstrations, pages 30–35, Uppsala, Sweden, July. Association for Computational Linguistics. Woosung Kim and Sanjeev Khudanpur. 2004. Crosslingual latent semantic analysis for language modeling. In IEEE international conference on acoustics, speech, and signal processing (ICASSP), volume 1, pages 257–260, Montr ´eal. Philipp Koehn, Franz Josef Och, and Daniel Marcu. 2003. Statistical phrase-based translation. In Proceedings of the 2003 conference of the North American chapter of the Association for Computational Linguistics on Human Language Technology, pages 48– 54, Edmonton. 1189 Philipp Koehn, Hieu Hoang, Alexandra Birch, et al. 2007. Moses: open source toolkit for Statistical Machine Translation. In Annual meeting of the Association for Computational Linguistics: Demonstration session, pages 177–180, Prague. Philippe Langlais, Alexandre Patry, and Fabrizio Gotti. 2007. A greedy decoder for phrase-based statistical machine translation. In TMI-2007: Proceedings of the 11th International Conference on Theoretical and Methodological Issues in Machine Translation, pages 104–1 13, Sk¨ ovde. Philippe Langlais, Alexandre Patry, and Fabrizio Gotti. 2008. Recherche locale pour la traduction statistique par segments. In TALN 2008, pages 119–128, Avignon, France, June. ATALA. Ronan Le Nagard and Philipp Koehn. 2010. Aiding pronoun translation with co-reference resolution. In Proceedings of the Joint Fifth Workshop on Statistical Machine Translation and MetricsMATR, pages 252–261, Uppsala, Sweden, July. Association for Computational Linguistics. Franz Josef Och, Nicola Ueffing, and Hermann Ney. 2001. An efficient A* search algorithm for Statistical Machine Translation. In Proceedings of the DataDriven Machine Translation Workshop, 39th Annual Meeting of the Association for Computational Linguistics (ACL), pages 55–62, Toulouse. Franz Josef Och. 2003. Minimum error rate training in Statistical Machine Translation. In Proceedings of the 41st annual meeting of the Association for Computational Linguistics, pages 160–167, Sapporo (Japan). Kishore Papineni, Salim Roukos, Todd Ward, and WeiJing Zhu. 2002. BLEU: a method for automatic evaluation of Machine Translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 3 11–3 18, Philadelphia. ACL. Yik-Cheung Tam, Ian Lane, and Tanja Schultz. 2007. Bilingual LSA-based adaptation for Statistical Machine Translation. Machine Translation, 21(4): 187– 207. J o¨rg Tiedemann. 2010. To cache or not to cache? Experiments with adaptive models in Statistical Machine Translation. In Proceedings of the ACL 2010 Joint Fifth Workshop on Statistical Machine Translation and Metrics MATR, pages 189–194, Uppsala, Sweden. Association for Computational Linguistics. Christoph Tillmann and Hermann Ney. 2003. Word reordering and a Dynamic Programming beam search algorithm for Statistical Machine Translation. Computational linguistics, 29(1):97–133. Christoph Tillmann, Stephan Vogel, Hermann Ney, and Alex Zubiaga. 1997. A DP-based search using monotone alignments in Statistical Translation. In Proceedings of the 35th Annual Meeting of the Association for Computational Spain, tics. Linguistics, July. Association for 289–296, Madrid, Computational Linguis- pages Tonio Wandmacher and Jean-Yves Antoine. 2007. Methods to integrate a language model with semantic information for a word prediction component. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLPCoNLL), pages 506–5 13, Prague, Czech Republic, June. Association for Computational Linguistics. 1190

5 0.13539718 42 emnlp-2012-Entropy-based Pruning for Phrase-based Machine Translation

Author: Wang Ling ; Joao Graca ; Isabel Trancoso ; Alan Black

Abstract: Phrase-based machine translation models have shown to yield better translations than Word-based models, since phrase pairs encode the contextual information that is needed for a more accurate translation. However, many phrase pairs do not encode any relevant context, which means that the translation event encoded in that phrase pair is led by smaller translation events that are independent from each other, and can be found on smaller phrase pairs, with little or no loss in translation accuracy. In this work, we propose a relative entropy model for translation models, that measures how likely a phrase pair encodes a translation event that is derivable using smaller translation events with similar probabilities. This model is then applied to phrase table pruning. Tests show that considerable amounts of phrase pairs can be excluded, without much impact on the transla- . tion quality. In fact, we show that better translations can be obtained using our pruned models, due to the compression of the search space during decoding.

6 0.11316119 135 emnlp-2012-Using Discourse Information for Paraphrase Extraction

7 0.1098384 86 emnlp-2012-Locally Training the Log-Linear Model for SMT

8 0.081786029 82 emnlp-2012-Left-to-Right Tree-to-String Decoding with Prediction

9 0.07796777 1 emnlp-2012-A Bayesian Model for Learning SCFGs with Discontiguous Rules

10 0.074325487 67 emnlp-2012-Inducing a Discriminative Parser to Optimize Machine Translation Reordering

11 0.062586099 2 emnlp-2012-A Beam-Search Decoder for Grammatical Error Correction

12 0.060282916 127 emnlp-2012-Transforming Trees to Improve Syntactic Convergence

13 0.058179565 18 emnlp-2012-An Empirical Investigation of Statistical Significance in NLP

14 0.056587774 109 emnlp-2012-Re-training Monolingual Parser Bilingually for Syntactic SMT

15 0.055100355 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction

16 0.054790128 54 emnlp-2012-Forced Derivation Tree based Model Training to Statistical Machine Translation

17 0.054097373 126 emnlp-2012-Training Factored PCFGs with Expectation Propagation

18 0.052771896 94 emnlp-2012-Multiple Aspect Summarization Using Integer Linear Programming

19 0.048001621 12 emnlp-2012-A Transition-Based System for Joint Part-of-Speech Tagging and Labeled Non-Projective Dependency Parsing

20 0.047543433 104 emnlp-2012-Parse, Price and Cut-Delayed Column and Row Generation for Graph Based Parsers

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.201), (1, -0.174), (2, -0.161), (3, -0.004), (4, -0.086), (5, -0.07), (6, -0.109), (7, 0.052), (8, 0.059), (9, 0.099), (10, 0.034), (11, -0.084), (12, 0.08), (13, -0.116), (14, 0.094), (15, -0.082), (16, -0.063), (17, -0.078), (18, -0.22), (19, -0.066), (20, 0.022), (21, 0.109), (22, 0.096), (23, 0.024), (24, -0.08), (25, -0.015), (26, 0.047), (27, 0.016), (28, 0.063), (29, 0.005), (30, -0.015), (31, -0.098), (32, 0.021), (33, -0.05), (34, 0.041), (35, 0.136), (36, 0.02), (37, 0.022), (38, 0.144), (39, -0.057), (40, 0.033), (41, 0.022), (42, 0.107), (43, 0.057), (44, -0.088), (45, 0.171), (46, -0.041), (47, -0.024), (48, 0.049), (49, 0.0)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.93952584 74 emnlp-2012-Language Model Rest Costs and Space-Efficient Storage

Author: Kenneth Heafield ; Philipp Koehn ; Alon Lavie

2 0.62091732 11 emnlp-2012-A Systematic Comparison of Phrase Table Pruning Techniques

Author: Richard Zens ; Daisy Stanton ; Peng Xu

3 0.57010472 42 emnlp-2012-Entropy-based Pruning for Phrase-based Machine Translation

Author: Wang Ling ; Joao Graca ; Isabel Trancoso ; Alan Black

4 0.53140807 57 emnlp-2012-Generalized Higher-Order Dependency Parsing with Cube Pruning

Author: Hao Zhang ; Ryan McDonald

5 0.50698805 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

6 0.44561464 86 emnlp-2012-Locally Training the Log-Linear Model for SMT

7 0.41417035 82 emnlp-2012-Left-to-Right Tree-to-String Decoding with Prediction

8 0.34056839 1 emnlp-2012-A Bayesian Model for Learning SCFGs with Discontiguous Rules

9 0.32610342 2 emnlp-2012-A Beam-Search Decoder for Grammatical Error Correction

10 0.31842738 18 emnlp-2012-An Empirical Investigation of Statistical Significance in NLP

11 0.31232798 118 emnlp-2012-Source Language Adaptation for Resource-Poor Machine Translation

12 0.30466816 26 emnlp-2012-Building a Lightweight Semantic Model for Unsupervised Information Extraction on Short Listings

13 0.29410523 108 emnlp-2012-Probabilistic Finite State Machines for Regression-based MT Evaluation

14 0.2774018 135 emnlp-2012-Using Discourse Information for Paraphrase Extraction

15 0.27059454 54 emnlp-2012-Forced Derivation Tree based Model Training to Statistical Machine Translation

16 0.26867262 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction

17 0.26494697 130 emnlp-2012-Unambiguity Regularization for Unsupervised Learning of Probabilistic Grammars

18 0.26031104 104 emnlp-2012-Parse, Price and Cut-Delayed Column and Row Generation for Graph Based Parsers

19 0.2553955 126 emnlp-2012-Training Factored PCFGs with Expectation Propagation

20 0.24001385 105 emnlp-2012-Parser Showdown at the Wall Street Corral: An Empirical Investigation of Error Types in Parser Output

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(2, 0.015), (14, 0.434), (16, 0.04), (25, 0.011), (34, 0.097), (41, 0.02), (60, 0.068), (63, 0.039), (65, 0.018), (70, 0.016), (74, 0.059), (76, 0.026), (79, 0.018), (80, 0.013), (86, 0.031), (95, 0.024)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.80103385 74 emnlp-2012-Language Model Rest Costs and Space-Efficient Storage

Author: Kenneth Heafield ; Philipp Koehn ; Alon Lavie

2 0.74971664 104 emnlp-2012-Parse, Price and Cut-Delayed Column and Row Generation for Graph Based Parsers

Author: Sebastian Riedel ; David Smith ; Andrew McCallum

Abstract: Graph-based dependency parsers suffer from the sheer number of higher order edges they need to (a) score and (b) consider during optimization. Here we show that when working with LP relaxations, large fractions of these edges can be pruned before they are fully scored—without any loss of optimality guarantees and, hence, accuracy. This is achieved by iteratively parsing with a subset of higherorder edges, adding higher-order edges that may improve the score of the current solution, and adding higher-order edges that are implied by the current best first order edges. This amounts to delayed column and row generation in the LP relaxation and is guaranteed to provide the optimal LP solution. For second order grandparent models, our method considers, or scores, no more than 6–13% of the second order edges of the full model. This yields up to an eightfold parsing speedup, while pro- viding the same empirical accuracy and certificates of optimality as working with the full LP relaxation. We also provide a tighter LP formulation for grandparent models that leads to a smaller integrality gap and higher speed.

3 0.35982457 11 emnlp-2012-A Systematic Comparison of Phrase Table Pruning Techniques

Author: Richard Zens ; Daisy Stanton ; Peng Xu

4 0.35236102 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

5 0.35005894 42 emnlp-2012-Entropy-based Pruning for Phrase-based Machine Translation

Author: Wang Ling ; Joao Graca ; Isabel Trancoso ; Alan Black

6 0.34375975 67 emnlp-2012-Inducing a Discriminative Parser to Optimize Machine Translation Reordering

7 0.33633098 18 emnlp-2012-An Empirical Investigation of Statistical Significance in NLP

8 0.3295711 75 emnlp-2012-Large Scale Decipherment for Out-of-Domain Machine Translation

9 0.32781798 108 emnlp-2012-Probabilistic Finite State Machines for Regression-based MT Evaluation

10 0.32497558 8 emnlp-2012-A Phrase-Discovering Topic Model Using Hierarchical Pitman-Yor Processes

11 0.32079223 52 emnlp-2012-Fast Large-Scale Approximate Graph Construction for NLP