emnlp emnlp2010 emnlp2010-19 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Erica Greene ; Tugba Bodrumlu ; Kevin Knight
Abstract: Tugba Bodrumlu Dept. of Computer Science Univ. of Southern California Los Angeles, CA 90089 bodrumlu@cs . usc . edu Kevin Knight Information Sciences Institute Univ. of Southern California 4676 Admiralty Way Marina del Rey, CA 90292 kn i @ i i ght s .edu from existing online poetry corpora. We use these patterns to generate new poems and translate exist- We employ statistical methods to analyze, generate, and translate rhythmic poetry. We first apply unsupervised learning to reveal word-stress patterns in a corpus of raw poetry. We then use these word-stress patterns, in addition to rhyme and discourse models, to generate English love poetry. Finally, we translate Italian poetry into English, choosing target realizations that conform to desired rhythmic patterns.
Reference: text
sentIndex sentText sentNum sentScore
1 We use these patterns to generate new poems and translate exist- We employ statistical methods to analyze, generate, and translate rhythmic poetry. [sent-10, score-0.369]
2 We then use these word-stress patterns, in addition to rhyme and discourse models, to generate English love poetry. [sent-12, score-0.251]
3 Finally, we translate Italian poetry into English, choosing target realizations that conform to desired rhythmic patterns. [sent-13, score-0.7]
4 1 Introduction When it comes to generating creative language (poems, stories, jokes, etc), people have massive advantages over machines: • people can construct grammatical, sensible utterances, • people have a wide range of topics to talk about, an hadv • people experience joy and heart-break. [sent-14, score-0.136]
5 In this paper we concentrate on statistical methods applied to the analysis, generation, and translation of poetry. [sent-16, score-0.085]
6 When translating, we render target text in a rhythmic scheme determined by the user. [sent-18, score-0.137]
7 Poetry generation has received research attention in the past (Manurung et al. [sent-19, score-0.07]
8 Less research effort has been spent on poetry analysis and poetry translation, which we tackle here. [sent-25, score-0.994]
9 2 Terms Meter refers to the rhythmic beat of poetic text when read aloud. [sent-26, score-0.22]
10 Iambic is a common meter that sounds like da-DUM da-DUM da-DUM, etc. [sent-27, score-0.124]
11 Trimeter refers to a line with three feet, pentameter to a line with five feet, etc. [sent-30, score-0.265]
12 (iambic pentameter) • twas the NIGHT before CHRIST-mas and tAwLaLs through HtheT H beOfUorSeE (anapest mteatsram anedter) Classical English sonnets are poems most often composed of 14 lines of iambic pentameter. [sent-32, score-0.692]
13 tc ho2d0s10 in A Nsastoucira tlio Lnan fogru Cagoem Ppruotcaetisosninagl, L pinag eusis 5t2ic4s–53 , 3 Analysis We focus on English rhythmic poetry. [sent-35, score-0.137]
14 We define the following analysis task: given poetic lines in a known meter (such as sonnets written in iambic pentameter), assign a syllable-stress pattern to each word in each line. [sent-36, score-0.75]
15 Making such decisions is part of the larger task of reading poetry aloud. [sent-37, score-0.497]
16 Later in the paper, we will employ the concrete statistical tables from analysis to the problems of poetry generation and translation. [sent-38, score-0.567]
17 We create a test set consisting of 70 lines from Shakespeare’s sonnets, which are written in iambic pentameter. [sent-39, score-0.472]
18 shal l i compare thee t o a summe r s day | | /\ | | | /\ | S S* S S* S S* S S* S S* S refers to an unstressed syllable, and S* refers to a stressed syllable. [sent-41, score-0.262]
19 One of the authors created goldstandard output by listening to Internet recordings of the 70 lines and marking words according to the speaker’s stress. [sent-42, score-0.158]
20 The task evaluation consists of perword accuracy (how many words are assigned the correct stress pattern) and per-line accuracy (how many lines have all words analyzed perfectly). [sent-43, score-0.183]
21 Even when all words are known, many lines dEov nnot w seem tlo w coorndtsai anr e1 k0n syllables. [sent-46, score-0.112]
22 Spoken recordings include stress reversals, sSupcohk as poin-TING i innstcelaudd eof s POIN-ting. [sent-48, score-0.117]
23 When we generate rhythmic text, it is important to use onesyllable words properly. [sent-53, score-0.177]
24 For example, we would be happy for an iambic generator to output big thoughts are not quite here, but not quite big thoughts are not here. [sent-54, score-0.44]
25 For raw data, we start with all Shakespeare sonnets (17,134 word tokens). [sent-58, score-0.12]
26 Figures 1 and 2 show a finite-state transducer (FST) that converts sequences of English words to sequences of S* and S symbols. [sent-60, score-0.188]
27 For example, the train-cascade command uses EM to learn probabilities in an arbitrary FST cascade from end-to-end input/output string pairs. [sent-72, score-0.16]
28 This machine maps sequences of English words onto sequences of SFi*g aurned 2S: s Aymn ebfoflicsi, representing estmreesnsteindg ga Pnd(m m un|es)t. [sent-74, score-0.148]
29 526 Figure 3: An FST that accepts any of four input meters and deterministically normalizes its input to strict iambic pentameter. [sent-78, score-0.488]
30 e → P(m|e) → m → norm → m Figure 4: FST cascade that encodes a loose interpretation of iambic pentameter. [sent-80, score-0.564]
31 The norm FST accepts any of four near-iambic-pentameter sequences and normalizes them into strict iambic pentameter. [sent-81, score-0.555]
32 Note that the output sequences are all the same, representing our belief that each line should be read as iambic After we train the FST, we can use Viterbi decoding to recover the highestprobability alignments, e. [sent-82, score-0.53]
33 First, we augment the Shakespeare sonnets with data from the website sonnets. [sent-88, score-0.12]
34 : from | S fai re st /\ S* creature s /\ S S* we de s i re increa s e /\ /\ | S S* S S* S S* Second, we loosen our model. [sent-93, score-0.432]
35 When we listen to recordings, we discover that not all lines are read S S* S S* S S* S S* S S*. [sent-94, score-0.161]
36 Indeed, some lines in our data contain eleven words—these are unexplainable by the EM training system. [sent-95, score-0.112]
37 We also observe that 2We can augment the data with lines of poetry written in meters other than iambic pentameter, so long as we supply the desired output pattern for each input line. [sent-96, score-1.02]
38 These variations yield four possible syllable-stress sequences: S S* S S* S S* S S* S S* S* S S S* S S* S S* S S* S S* S S* S S* S S* S S* S S* S S S* S S* S S* S S* S We want to offer EM the freedom to analyze lines into any of these four variations. [sent-106, score-0.112]
39 We therefore construct a second FST (Figure 3), norm, which maps all four sequences onto the canonical pattern S S* S S* S S* S S* S S*. [sent-107, score-0.074]
40 We then arrange both FSTs in a cascade (Figure 4), and we train the whole cascade on the same input/output sequences as before. [sent-108, score-0.394]
41 Viterbi decoding through the two-step cascade now reveals EM’s proposed internal meter analysis as well as token mappings, e. [sent-110, score-0.284]
42 81% a syllable- the pattern transcribed of whole lines are also The upper limit for whole-line under our constraints 88. [sent-116, score-0.112]
43 We further obtain a probabilistic mappings table of word that we can use for generation and trans- P (S* S S* P (S* S P (S* S P (S S* | creature s ) | point ed) | point ed) P (S* S P (S* S S* P (S* S P (S* | alt itude ) | pri s oner ) | pri s oner ) | mothe r ) | mothe r ) = 1. [sent-118, score-0.382]
44 W* e| mwootuhlder incorrectly le smarnal a mroubcahhigher value if we did not loosen the iambic pentameter model, as many mother tokens occur lineinitial and line-final. [sent-133, score-0.617]
45 Figure 7 shows which one-syllable words are more often stressed (or unstressed) in iambic pentameter poetry. [sent-134, score-0.624]
46 Alignment errors still occur, especially in noisy 528 P(m) → m → P(e|m) → e → P(e) → e Figure 8: Finite-state cascade for poetry generation. [sent-137, score-0.657]
47 portions of the data that are not actually written in iambic pentameter, but also in clean portions, e. [sent-138, score-0.36]
48 : the | pe r fect /\ ceremony / |\ of | love s | rit e /\ S S* S S* S S* S S* S S* The word ceremony only occurs this once in the data, so it is willing to accept any stress pattern. [sent-140, score-0.219]
49 While rite is correctly analyzed elsewhere as a onesyllable word, loves prefers S*, and this overwhelms the one-syllable preference for rite. [sent-141, score-0.08]
50 We can blame our tokenizer for this, as it conflates loves and love’s, despite the fact that these words have different stress probabilities. [sent-142, score-0.111]
51 4 Generation Figure 8 shows our concept of generation as a cascade of weighted FSTs. [sent-143, score-0.23]
52 Finally, P(e) is a word-trigram model built from a 10,000-line corpus of 105 English love poems. [sent-154, score-0.08]
53 We select the first line of our poem from the FST cascade’s 100,000-best list, or by hand. [sent-155, score-0.081]
54 To generate each subsequent line, we modify the cascade and run it again. [sent-156, score-0.16]
55 From our poetry corpus, we estimate a word’s unigram probability given the words on the previous line, via IBM Model 1 (Brown et al. [sent-158, score-0.497]
56 Second, we check if any previous line The women of the night Again and al l the way Li ke a mou s e in the whit e Not . [sent-161, score-0.127]
57 - Of the heart of the day - the bed to t rust Around her twi st s me the st i wi l not t e l thee l l Fi re change s everything . [sent-162, score-0.264]
58 If so, we build an additional FST that accepts only strings whose final word rhymes with wn. [sent-165, score-0.077]
59 This is a reasonable approach, though it will not, for example, rhyme . [sent-166, score-0.171]
60 We say two non-identical words rhyme if their phoneme strings share a common suffix that includes the last stressed vowel. [sent-173, score-0.264]
61 wn Figure 9 shows several poems that we automatically generate with this scheme. [sent-177, score-0.1]
62 5 Translation Automatically generated poetry can sound good when read aloud, but it often has a “nonsense” feel to it. [sent-178, score-0.546]
63 Translation provides one way to tie things to529 i → P(e|i) → e → P(m|e) → m → P(m) → m Figure 10: Finite-state cascade for poetry translation. [sent-180, score-0.72]
64 For example, we may want to translate Italian sonnets into fluent English iambic pentameter. [sent-183, score-0.546]
65 The poem begins: ne l me z z o de l cammin di no st ra vit a mi ritrovai per una selva oscura che la via diritt a era smarrit a . [sent-186, score-0.124]
66 The meter in Italian is hendecasyllabic, which has ten syllables and ensures three beats. [sent-188, score-0.167]
67 Dante’s Italian rhyme scheme is: ABA, BCB, CDC, etc, meaning that lines 2, 4, and 6 rhyme with each other; lines 5, 7, and 9 rhyme with each other, and so forth. [sent-189, score-0.737]
68 Some translations target iambic pentameter, but even the most respected translations give up on rhyme, since English is much harder to rhyme than Italian. [sent-194, score-0.645]
69 Longfellow’s translation begins: midway upon the j ourney o f our l fe i i found mys e l within a fore st dark f for the st raight forward pathway had been lo st . [sent-195, score-0.746]
70 We arrange the translation problem as a cascade of WFSTs, as shown in Figure 10. [sent-196, score-0.245]
71 In lieu of the first WFST, we use the statistical phrase-based machine translation (PBMT) system Moses (Koehn et al. [sent-198, score-0.085]
72 Finally, we efi Plte(mr t|hee) dreesvuilcteing syllable sequences with a strict, single-path, deterministic iambic pentameter acceptor, P(m). [sent-202, score-0.667]
73 3 Our 3It is also possible to use a looser iambic P(m) model, as described in Section 3. [sent-203, score-0.36]
74 finite-state toolkit’s top-k paths represent the translations with the highest product of scores P(e|i) · P(m|e) · P(m). [sent-211, score-0.093]
75 This creates output lines that do not scan easily. [sent-214, score-0.16]
76 The vast majority of parallel Italian/English poetry is DC itself, for which we have four English translations. [sent-218, score-0.497]
77 We augment our target language model with English poetry collected from many sources. [sent-220, score-0.497]
78 Setting the weight sto ion high (rems|uelt)s mino dlienles to oth tahet scan very well, S beutt nwgho thsee translation quality is low. [sent-226, score-0.133]
79 In this test-on-train scenario, the machine reproduces lines from human translations it has seen. [sent-228, score-0.169]
80 This is a normal scenario in human poetry translation, where people have access to previous translations. [sent-231, score-0.531]
81 Figure 12 shows how we translate the first lines of DC, first using only PBMT, then using the full system. [sent-232, score-0.178]
82 : mi dway upon the j ourney /\ /\ | /\ S S* S S* S S* S The machine’s Longfellow’s, translation the system’s intended o f our l fe i | | | S* S S* here is the same as which is in the training data. [sent-235, score-0.225]
83 To obtain iambic tetrameter (4-beat) translations, we delete the last two transitions of the P(m) model. [sent-240, score-0.36]
84 We then get: I : in our l fe the j ourney way 4 i i found mys e l deep on dark wood f that lo st st raight forward pathway had . [sent-241, score-0.58]
85 ah how t o s ay the what i hard s thi s fore st s avage rough and st ern the very thought renews the fear . [sent-242, score-0.231]
86 The third line fails because all paths through the translation lattice contain an A somewhere. [sent-245, score-0.168]
87 Figure 13 shows a portion of the translation results. [sent-248, score-0.085]
88 The P(m|e) meter model cannot process those words, accounting efoterr rth meo Id5e lfai claunren o ratt ep. [sent-250, score-0.124]
89 r Here, we get a first look at statistical MT translating poetry into rhythmic structures—as with all MT, there are successes and problems, and certainly more to do. [sent-251, score-0.675]
90 and translation do not always scan naturally when read aloud by a person. [sent-258, score-0.222]
91 Developing a model with contextdependent probabilities may be useful not only for improving generation and translation, but also for improving poetry analysis itself, as measured by anlaysis task accuracy. [sent-265, score-0.567]
92 Evaluation is a big open problem for automatic poetry generation—even evaluating human poetry is difficult. [sent-268, score-0.994]
93 , confounding judges attempts to distinguish machine poetry from human poetry. [sent-271, score-0.497]
94 The advantage of translation over generation is that the source text provides a coherent sequence of propositions and images, allowing the machine to focus on “how to say” instead of “what to say. [sent-274, score-0.155]
95 ” However, translation output lattices offer limited material to work with, and as we dig deeper into those lattices, we encounter increasingly disfluent ways to string together renderings of the source substrings. [sent-275, score-0.085]
96 An appealing future direction is to combine translation and generation. [sent-276, score-0.085]
97 Such a hybrid translation/generation program would not be bound to translate every word, but rather it could more freely combine lexical material from its translation tables 532 with other grammatical and lexical resources. [sent-278, score-0.151]
98 Interestingly, human translators sometimes work this way when they translate poetry—many excellent works have been produced by people with very little knowledge of the source language. [sent-279, score-0.1]
99 Recently, e→f translation tables haPvea baepehnr composed ewntilthy, fe→→ef tables, oton tmabalekse e→e ta bebleens t choamt can paraphrase English isn,to t English (Bannard asn tdh Callison-Burch, 2005). [sent-281, score-0.085]
100 hT ihnisto om Eankgeliss iht possible to consider statistical translation of English prose into English poetry. [sent-282, score-0.119]
wordName wordTfidf (topN-words)
[('poetry', 0.497), ('iambic', 0.36), ('fst', 0.32), ('pentameter', 0.171), ('rhyme', 0.171), ('cascade', 0.16), ('rhythmic', 0.137), ('meter', 0.124), ('sonnets', 0.12), ('lines', 0.112), ('poems', 0.1), ('italian', 0.093), ('stressed', 0.093), ('dc', 0.091), ('st', 0.09), ('unstressed', 0.086), ('translation', 0.085), ('creature', 0.08), ('fai', 0.08), ('love', 0.08), ('ourney', 0.08), ('sequences', 0.074), ('stress', 0.071), ('generation', 0.07), ('haiku', 0.069), ('translate', 0.066), ('aw', 0.063), ('things', 0.063), ('syllable', 0.062), ('beast', 0.06), ('dark', 0.06), ('divine', 0.06), ('gervas', 0.06), ('hou', 0.06), ('increa', 0.06), ('mothe', 0.06), ('mys', 0.06), ('pbmt', 0.06), ('wood', 0.06), ('fe', 0.06), ('translations', 0.057), ('manurung', 0.051), ('meters', 0.051), ('shakespeare', 0.051), ('dante', 0.051), ('eye', 0.051), ('fore', 0.051), ('pronunciation', 0.05), ('read', 0.049), ('scan', 0.048), ('line', 0.047), ('mother', 0.046), ('recordings', 0.046), ('norm', 0.044), ('mappings', 0.044), ('syllables', 0.043), ('accepts', 0.043), ('thee', 0.043), ('re', 0.041), ('translating', 0.041), ('aloud', 0.04), ('anapest', 0.04), ('bodrumlu', 0.04), ('bright', 0.04), ('gravitate', 0.04), ('haverford', 0.04), ('longfellow', 0.04), ('loosen', 0.04), ('loves', 0.04), ('mou', 0.04), ('netzer', 0.04), ('night', 0.04), ('onesyllable', 0.04), ('pathway', 0.04), ('raight', 0.04), ('shal', 0.04), ('thine', 0.04), ('thou', 0.04), ('thoughts', 0.04), ('tosa', 0.04), ('transducer', 0.04), ('trimeter', 0.04), ('int', 0.04), ('paths', 0.036), ('ld', 0.036), ('people', 0.034), ('english', 0.034), ('feet', 0.034), ('oner', 0.034), ('poem', 0.034), ('poetic', 0.034), ('prose', 0.034), ('rhymes', 0.034), ('fir', 0.034), ('carmel', 0.034), ('ceremony', 0.034), ('ime', 0.034), ('moon', 0.034), ('normalizes', 0.034), ('em', 0.034)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999809 19 emnlp-2010-Automatic Analysis of Rhythmic Poetry with Applications to Generation and Translation
Author: Erica Greene ; Tugba Bodrumlu ; Kevin Knight
Abstract: Tugba Bodrumlu Dept. of Computer Science Univ. of Southern California Los Angeles, CA 90089 bodrumlu@cs . usc . edu Kevin Knight Information Sciences Institute Univ. of Southern California 4676 Admiralty Way Marina del Rey, CA 90292 kn i @ i i ght s .edu from existing online poetry corpora. We use these patterns to generate new poems and translate exist- We employ statistical methods to analyze, generate, and translate rhythmic poetry. We first apply unsupervised learning to reveal word-stress patterns in a corpus of raw poetry. We then use these word-stress patterns, in addition to rhyme and discourse models, to generate English love poetry. Finally, we translate Italian poetry into English, choosing target realizations that conform to desired rhythmic patterns.
2 0.40589783 1 emnlp-2010-"Poetic" Statistical Machine Translation: Rhyme and Meter
Author: Dmitriy Genzel ; Jakob Uszkoreit ; Franz Och
Abstract: As a prerequisite to translation of poetry, we implement the ability to produce translations with meter and rhyme for phrase-based MT, examine whether the hypothesis space of such a system is flexible enough to accomodate such constraints, and investigate the impact of such constraints on translation quality.
3 0.090947233 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
Author: Lei Shi ; Rada Mihalcea ; Mingjun Tian
Abstract: In this paper, we introduce a method that automatically builds text classifiers in a new language by training on already labeled data in another language. Our method transfers the classification knowledge across languages by translating the model features and by using an Expectation Maximization (EM) algorithm that naturally takes into account the ambiguity associated with the translation of a word. We further exploit the readily available unlabeled data in the target language via semisupervised learning, and adapt the translated model to better fit the data distribution of the target language.
4 0.07385201 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
Author: Philip Resnik ; Olivia Buzek ; Chang Hu ; Yakov Kronrod ; Alex Quinn ; Benjamin B. Bederson
Abstract: Targeted paraphrasing is a new approach to the problem of obtaining cost-effective, reasonable quality translation that makes use of simple and inexpensive human computations by monolingual speakers in combination with machine translation. The key insight behind the process is that it is possible to spot likely translation errors with only monolingual knowledge of the target language, and it is possible to generate alternative ways to say the same thing (i.e. paraphrases) with only monolingual knowledge of the source language. Evaluations demonstrate that this approach can yield substantial improvements in translation quality.
5 0.07344313 47 emnlp-2010-Example-Based Paraphrasing for Improved Phrase-Based Statistical Machine Translation
Author: Aurelien Max
Abstract: In this article, an original view on how to improve phrase translation estimates is proposed. This proposal is grounded on two main ideas: first, that appropriate examples of a given phrase should participate more in building its translation distribution; second, that paraphrases can be used to better estimate this distribution. Initial experiments provide evidence of the potential of our approach and its implementation for effectively improving translation performance.
6 0.072990686 50 emnlp-2010-Facilitating Translation Using Source Language Paraphrase Lattices
7 0.067251533 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
8 0.061327048 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
9 0.049852811 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
10 0.049180232 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
11 0.048785865 5 emnlp-2010-A Hybrid Morpheme-Word Representation for Machine Translation of Morphologically Rich Languages
12 0.047886726 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
13 0.045939118 22 emnlp-2010-Automatic Evaluation of Translation Quality for Distant Language Pairs
14 0.041343749 104 emnlp-2010-The Necessity of Combining Adaptation Methods
15 0.03910476 39 emnlp-2010-EMNLP 044
16 0.036670972 99 emnlp-2010-Statistical Machine Translation with a Factorized Grammar
17 0.036537316 29 emnlp-2010-Combining Unsupervised and Supervised Alignments for MT: An Empirical Study
18 0.036291402 98 emnlp-2010-Soft Syntactic Constraints for Hierarchical Phrase-Based Translation Using Latent Syntactic Distributions
19 0.035383623 62 emnlp-2010-Improving Mention Detection Robustness to Noisy Input
20 0.03499594 36 emnlp-2010-Discriminative Word Alignment with a Function Word Reordering Model
topicId topicWeight
[(0, 0.161), (1, -0.112), (2, -0.053), (3, 0.024), (4, -0.053), (5, 0.043), (6, -0.01), (7, 0.014), (8, -0.126), (9, -0.065), (10, 0.044), (11, -0.016), (12, 0.074), (13, -0.036), (14, 0.752), (15, -0.043), (16, -0.012), (17, -0.003), (18, 0.075), (19, -0.009), (20, 0.087), (21, -0.037), (22, 0.029), (23, -0.161), (24, 0.011), (25, 0.016), (26, -0.082), (27, 0.034), (28, -0.035), (29, 0.025), (30, 0.019), (31, -0.004), (32, -0.004), (33, -0.021), (34, -0.001), (35, -0.034), (36, -0.001), (37, 0.013), (38, -0.052), (39, 0.015), (40, 0.009), (41, 0.003), (42, 0.006), (43, 0.031), (44, 0.02), (45, -0.003), (46, 0.014), (47, 0.027), (48, -0.02), (49, -0.018)]
simIndex simValue paperId paperTitle
same-paper 1 0.93722707 19 emnlp-2010-Automatic Analysis of Rhythmic Poetry with Applications to Generation and Translation
Author: Erica Greene ; Tugba Bodrumlu ; Kevin Knight
Abstract: Tugba Bodrumlu Dept. of Computer Science Univ. of Southern California Los Angeles, CA 90089 bodrumlu@cs . usc . edu Kevin Knight Information Sciences Institute Univ. of Southern California 4676 Admiralty Way Marina del Rey, CA 90292 kn i @ i i ght s .edu from existing online poetry corpora. We use these patterns to generate new poems and translate exist- We employ statistical methods to analyze, generate, and translate rhythmic poetry. We first apply unsupervised learning to reveal word-stress patterns in a corpus of raw poetry. We then use these word-stress patterns, in addition to rhyme and discourse models, to generate English love poetry. Finally, we translate Italian poetry into English, choosing target realizations that conform to desired rhythmic patterns.
2 0.89125639 1 emnlp-2010-"Poetic" Statistical Machine Translation: Rhyme and Meter
Author: Dmitriy Genzel ; Jakob Uszkoreit ; Franz Och
Abstract: As a prerequisite to translation of poetry, we implement the ability to produce translations with meter and rhyme for phrase-based MT, examine whether the hypothesis space of such a system is flexible enough to accomodate such constraints, and investigate the impact of such constraints on translation quality.
3 0.20489353 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
Author: Philip Resnik ; Olivia Buzek ; Chang Hu ; Yakov Kronrod ; Alex Quinn ; Benjamin B. Bederson
Abstract: Targeted paraphrasing is a new approach to the problem of obtaining cost-effective, reasonable quality translation that makes use of simple and inexpensive human computations by monolingual speakers in combination with machine translation. The key insight behind the process is that it is possible to spot likely translation errors with only monolingual knowledge of the target language, and it is possible to generate alternative ways to say the same thing (i.e. paraphrases) with only monolingual knowledge of the source language. Evaluations demonstrate that this approach can yield substantial improvements in translation quality.
4 0.19134325 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
Author: Guillaume Wisniewski ; Alexandre Allauzen ; Francois Yvon
Abstract: Extant Statistical Machine Translation (SMT) systems are very complex softwares, which embed multiple layers of heuristics and embark very large numbers of numerical parameters. As a result, it is difficult to analyze output translations and there is a real need for tools that could help developers to better understand the various causes of errors. In this study, we make a step in that direction and present an attempt to evaluate the quality of the phrase-based translation model. In order to identify those translation errors that stem from deficiencies in the phrase table (PT), we propose to compute the oracle BLEU-4 score, that is the best score that a system based on this PT can achieve on a reference corpus. By casting the computation of the oracle BLEU-1 as an Integer Linear Programming (ILP) problem, we show that it is possible to efficiently compute accurate lower-bounds of this score, and report measures performed on several standard benchmarks. Various other applications of these oracle decoding techniques are also reported and discussed. 1 Phrase-Based Machine Translation 1.1 Principle A Phrase-Based Translation System (PBTS) consists of a ruleset and a scoring function (Lopez, 2009). The ruleset, represented in the phrase table, is a set of phrase1pairs {(f, e) }, each pair expressing that the source phrase f can ,bee) r}e,w earicthten p (atirra enxslparteedss)i inngto t a target phrase e. Trarsaens flation hypotheses are generated by iteratively rewriting portions of the source sentence as prescribed by the ruleset, until each source word has been consumed by exactly one rule. The order of target words in an hypothesis is uniquely determined by the order in which the rewrite operation are performed. The search space ofthe translation model corresponds to the set of all possible sequences of 1Following the usage in statistical machine translation literature, use “phrase” to denote a subsequence of consecutive words. we 933 rules applications. The scoring function aims to rank all possible translation hypotheses in such a way that the best one has the highest score. A PBTS is learned from a parallel corpus in two independent steps. In a first step, the corpus is aligned at the word level, by using alignment tools such as Gi z a++ (Och and Ney, 2003) and some symmetrisation heuristics; phrases are then extracted by other heuristics (Koehn et al., 2003) and assigned numerical weights. In the second step, the parameters of the scoring function are estimated, typically through Minimum Error Rate training (Och, 2003). Translating a sentence amounts to finding the best scoring translation hypothesis in the search space. Because of the combinatorial nature of this problem, translation has to rely on heuristic search techniques such as greedy hill-climbing (Germann, 2003) or variants of best-first search like multi-stack decoding (Koehn, 2004). Moreover, to reduce the overall complexity of decoding, the search space is typically pruned using simple heuristics. For instance, the state-of-the-art phrase-based decoder Moses (Koehn et al., 2007) considers only a restricted number of translations for each source sequence2 and enforces a distortion limit3 over which phrases can be reordered. As a consequence, the best translation hypothesis returned by the decoder is not always the one with the highest score. 1.2 Typology of PBTS Errors Analyzing the errors of a SMT system is not an easy task, because of the number of models that are combined, the size of these models, and the high complexity of the various decision making processes. For a SMT system, three different kinds of errors can be distinguished (Germann et al., 2004; Auli et al., 2009): search errors, induction errors and model errors. The former corresponds to cases where the hypothesis with the best score is missed by the search procedure, either because of the use of an ap2the 3the option of Moses, defaulting to 20. dl option of Moses, whose default value is 7. tt l ProceMedITin,g Ms oasfs thaceh 2u0se1t0ts C,o UnSfAer,e n9c-e11 on O Ectmobpeir ic 2a0l1 M0.e ?tc ho2d0s10 in A Nsastouciraatlio Lnan fogru Cagoem Ppruotcaetisosninagl, L pinaggeusis 9t3ic3s–943, proximate search method or because of the restrictions of the search space. Induction errors correspond to cases where, given the model, the search space does not contain the reference. Finally, model errors correspond to cases where the hypothesis with the highest score is not the best translation according to the evaluation metric. Model errors encompass several types oferrors that occur during learning (Bottou and Bousquet, 2008)4. Approximation errors are errors caused by the use of a restricted and oversimplistic class of functions (here, finitestate transducers to model the generation of hypotheses and a linear scoring function to discriminate them) to model the translation process. Estimation errors correspond to the use of sub-optimal values for both the phrase pairs weights and the parameters of the scoring function. The reasons behind these errors are twofold: first, training only considers a finite sample of data; second, it relies on error prone alignments. As a result, some “good” phrases are extracted with a small weight, or, in the limit, are not extracted at all; and conversely that some “poor” phrases are inserted into the phrase table, sometimes with a really optimistic score. Sorting out and assessing the impact of these various causes of errors is of primary interest for SMT system developers: for lack of such diagnoses, it is difficult to figure out which components of the system require the most urgent attention. Diagnoses are however, given the tight intertwining among the various component of a system, very difficult to obtain: most evaluations are limited to the computation of global scores and usually do not imply any kind of failure analysis. 1.3 Contribution and organization To systematically assess the impact of the multiple heuristic decisions made during training and decoding, we propose, following (Dreyer et al., 2007; Auli et al., 2009), to work out oracle scores, that is to evaluate the best achievable performances of a PBTS. We aim at both studying the expressive power of PBTS and at providing tools for identifying and quantifying causes of failure. Under standard metrics such as BLEU (Papineni et al., 2002), oracle scores are difficult (if not impossible) to compute, but, by casting the computation of the oracle unigram recall and precision as an Integer Linear Programming (ILP) problem, we show that it is possible to efficiently compute accurate lower-bounds of the oracle BLEU-4 scores and report measurements performed on several standard benchmarks. The main contributions of this paper are twofold. We first introduce an ILP program able to efficiently find the best hypothesis a PBTS can achieve. This program can be easily extended to test various improvements to 4We omit here optimization errors. 934 phrase-base systems or to evaluate the impact of different parameter settings. Second, we present a number of complementary results illustrating the usage of our oracle decoder for identifying and analyzing PBTS errors. Our experimental results confirm the main conclusions of (Turchi et al., 2008), showing that extant PBTs have the potential to generate hypotheses having very high BLEU4 score and that their main bottleneck is their scoring function. The rest of this paper is organized as follows: in Section 2, we introduce and formalize the oracle decoding problem, and present a series of ILP problems of increasing complexity designed so as to deliver accurate lowerbounds of oracle score. This section closes with various extensions allowing to model supplementary constraints, most notably reordering constraints (Section 2.5). Our experiments are reported in Section 3, where we first introduce the training and test corpora, along with a description of our system building pipeline (Section 3. 1). We then discuss the baseline oracle BLEU scores (Section 3.2), analyze the non-reachable parts of the reference translations, and comment several complementary results which allow to identify causes of failures. Section 4 discuss our approach and findings with respect to the existing literature on error analysis and oracle decoding. We conclude and discuss further prospects in Section 5. 2 Oracle Decoder 2.1 The Oracle Decoding Problem Definition To get some insights on the errors of phrasebased systems and better understand their limits, we propose to consider the oracle decoding problem defined as follows: given a source sentence, its reference translation5 and a phrase table, what is the “best” translation hypothesis a system can generate? As usual, the quality of an hypothesis is evaluated by the similarity between the reference and the hypothesis. Note that in the oracle decoding problem, we are only assessing the ability of PBT systems to generate good candidate translations, irrespective of their ability to score them properly. We believe that studying this problem is interesting for various reasons. First, as described in Section 3.4, comparing the best hypothesis a system could have generated and the hypothesis it actually generates allows us to carry on both quantitative and qualitative failure analysis. The oracle decoding problem can also be used to assess the expressive power of phrase-based systems (Auli et al., 2009). Other applications include computing acceptable pseudo-references for discriminative training (Tillmann and Zhang, 2006; Liang et al., 2006; Arun and 5The oracle decoding problem can be extended to the case of multiple references. For the sake of simplicity, we only describe the case of a single reference. Koehn, 2007) or combining machine translation systems in a multi-source setting (Li and Khudanpur, 2009). We have also used oracle decoding to identify erroneous or difficult to translate references (Section 3.3). Evaluation Measure To fully define the oracle decoding problem, a measure of the similarity between a translation hypothesis and its reference translation has to be chosen. The most obvious choice is the BLEU-4 score (Papineni et al., 2002) used in most machine translation evaluations. However, using this metric in the oracle decoding problem raises several issues. First, BLEU-4 is a metric defined at the corpus level and is hard to interpret at the sentence level. More importantly, BLEU-4 is not decomposable6: as it relies on 4-grams statistics, the contribution of each phrase pair to the global score depends on the translation of the previous and following phrases and can not be evaluated in isolation. Because of its nondecomposability, maximizing BLEU-4 is hard; in particular, the phrase-level decomposability of the evaluation × metric is necessary in our approach. To circumvent this difficulty, we propose to evaluate the similarity between a translation hypothesis and a reference by the number of their common words. This amounts to evaluating translation quality in terms of unigram precision and recall, which are highly correlated with human judgements (Lavie et al., ). This measure is closely related to the BLEU-1 evaluation metric and the Meteor (Banerjee and Lavie, 2005) metric (when it is evaluated without considering near-matches and the distortion penalty). We also believe that hypotheses that maximize the unigram precision and recall at the sentence level yield corpus level BLEU-4 scores close the maximal achievable. Indeed, in the setting we will introduce in the next section, BLEU-1 and BLEU-4 are highly correlated: as all correct words of the hypothesis will be compelled to be at their correct position, any hypothesis with a high 1-gram precision is also bound to have a high 2-gram precision, etc. 2.2 Formalizing the Oracle Decoding Problem The oracle decoding problem has already been considered in the case of word-based models, in which all translation units are bound to contain only one word. The problem can then be solved by a bipartite graph matching algorithm (Leusch et al., 2008): given a n m binary matarligxo describing possible t 2r0an08sl)a:ti goinv elinn aks n b×emtw beeinna source words and target words7, this algorithm finds the subset of links maximizing the number of words of the reference that have been translated, while ensuring that each word 6Neither at the sentence (Chiang et al., 2008), nor at the phrase level. 7The (i, j) entry of the matrix is 1if the ith word of the source can be translated by the jth word of the reference, 0 otherwise. 935 is translated only once. Generalizing this approach to phrase-based systems amounts to solving the following problem: given a set of possible translation links between potential phrases of the source and of the target, find the subset of links so that the unigram precision and recall are the highest possible. The corresponding oracle hypothesis can then be easily generated by selecting the target phrases that are aligned with one source phrase, disregarding the others. In addition, to mimic the way OOVs are usually handled, we match identical OOV tokens appearing both in the source and target sentences. In this approach, the unigram precision is always one (every word generated in the oracle hypothesis matches exactly one word in the reference). As a consequence, to find the oracle hypothesis, we just have to maximize the recall, that is the number of words appearing both in the hypothesis and in the reference. Considering phrases instead of isolated words has a major impact on the computational complexity: in this new setting, the optimal segmentations in phrases of both the source and of the target have to be worked out in addition to links selection. Moreover, constraints have to be taken into account so as to enforce a proper segmentation of the source and target sentences. These constraints make it impossible to use the approach of (Leusch et al., 2008) and concur in making the oracle decoding problem for phrase-based models more complex than it is for word-based models: it can be proven, using arguments borrowed from (De Nero and Klein, 2008), that this problem is NP-hard even for the simple unigram precision measure. 2.3 An Integer Program for Oracle Decoding To solve the combinatorial problem introduced in the previous section, we propose to cast it into an Integer Linear Programming (ILP) problem, for which many generic solvers exist. ILP has already been used in SMT to find the optimal translation for word-based (Germann et al., 2001) and to study the complexity of learning phrase alignments (De Nero and Klein, 2008) models. Following the latter reference, we introduce the following variables: fi,j (resp. ek,l) is a binary indicator variable that is true when the phrase contains all spans from betweenword position i to j (resp. k to l) of the source (resp. target) sentence. We also introduce a binary variable, denoted ai,j,k,l, to describe a possible link between source phrase fi,j and target phrase ek,l. These variables are built from the entries of the phrase table according to selection strategies introduced in Section 2.4. In the following, index variables are so that: 0 ≤ i< j ≤ n, in the source sentence and 0 ≤ k < l ≤ m, in the target sentence, where n (resp. m) is the length of the source (resp. target) sentence. Solving the oracle decoding problem then amounts to optimizing the following objective function: mi,j,akx,li,Xj,k,lai,j,k,l· (l − k), (1) under the constraints: X ∀x ∈ J1,mK : ek,l ≤ 1 (2) = (3) 1∀,kn,lK : Xai,j,k,l = fk,l (4) ∀i,j : Xai,j,k,l (5) k,l s.tX. Xk≤x≤l ∀∀xy ∈∈ J11,,mnKK : X i,j s.tX. Xi≤y≤j fi,j 1 Xi,j = ei,j Xk,l The objective function (1) corresponds to the number of target words that are generated. The first set of constraints (2) ensures that each word in the reference e ap- pears in no more than one phrase. Maximizing the objective under these constraints amounts to maximizing the unigram recall. The second set of constraints (3) ensures that each word in the source f is translated exactly once, which guarantees that the search space of the ILP problem is the same as the search space of a phrase-based system. Constraints (4) bind the fk,l and ai,j,k,l variables, ensuring that whenever a link ai,j,k,l is active, the corresponding phrase fk,l is also active. Constraints (5) play a similar role for the reference. The Relaxed Problem Even though it accurately models the search space of a phrase-based decoder, this programs is not really useful as is: due to out-ofvocabulary words or missing entries in the phrase table, the constraint that all source words should be translated yields infeasible problems8. We propose to relax this problem and allow some source words to remain untranslated. This is done by replacing constraints (3) by: ∀y ∈ J1,nK : X i,j s.tX. Xi≤y≤j fi,j ≤ 1 To better ref∀lyec ∈t th J1e, bneKh :avior of phrase-based decoders, which attempt to translate all source words, we also need to modify the objective function as follows: X i,Xj,k,l ai,j,k,l · (l − k) +Xfi,j · (j − i) Xi,j (6) The second term in this new objective ensures that optimal solutions translate as many source words as possible. 8An ILP problem is said to be infeasible when tion violates at least one constraint. every possible solu- 936 The Relaxed-Distortion Problem A last caveat with the Relaxed optimization program is caused by frequently occurring source tokens, such as function words or punctuation signs, which can often align with more than one target word. For lack of taking distortion information into account in our objective function, all these alignments are deemed equivalent, even if some of them are clearly more satisfactory than others. This situation is illustrated on Figure 1. le chat et the cat and le the chien dog Figure 1: Equivalent alignments between “le” and “the”. The dashed lines corresponds to a less interpretable solution. To overcome this difficulty, we propose a last change to the objective function: X i,Xj,k,l ai,j,k,l · (l − k) +Xfi,j · (j − i) X ai,j,k,l|k − i| Xi,j −α (7) i Xk ,l X,j, Compared to the objective function of the relaxed problem (6), we introduce here a supplementary penalty factor which favors monotonous alignments. For each phrase pair, the higher the difference between source and target positions, the higher this penalty. If α is small enough, this extra term allows us to select, among all the optimal alignments of the re l axed problem, the one with the lowest distortion. In our experiments, we set α to min {n, m} to ensure that the penalty factor is always smminall{enr, ,tmha}n tthoe e rneswuarred t fhoart aligning atwltyo single iwso ardlwsa. 2.4 Selecting Indicator Variables In the approach introduced in the previous sections, the oracle decoding problem is solved by selecting, among a set of possible translation links, the ones that yield the solution with the highest unigram recall. We propose two strategies to build this set of possible translation links. In the first one, denoted exact match, an indicator ai,j,k,l is created if there is an entry (f, e) so that f spans from word position ito j in the source and e from word position k to l in the target. In this strategy, the ILP program considers exactly the same ruleset as conventional phrase-based decoders. We also consider an alternative strategy, which could help us to identify errors made during the phrase extraction process. In this strategy, denoted inside match, an indicator ai,j,k,l is created when the following three criteria are met: i) f spans from position ito j of the source; ii) a substring of e, denoted e, spans from position k to l of the reference; iii) (f, e¯) is not an entry of the phrase table. The resulting set of indicator variables thus contains, at least, all the variables used in the exact match strategy. In addition, we license here the use of phrases containing words that do not occur in the reference. In fact, using such solutions can yield higher BLEU scores when the reward for additional correct matches exceeds the cost incurred by wrong predictions. These cases are symptoms of situations where the extraction heuristic failed to extract potentially useful subphrases. 2.5 Oracle Decoding with Reordering Constraints The ILP problem introduced in the previous section can be extended in several ways to describe and test various improvements to phrase-based systems or to evaluate the impact of different parameter settings. This flexibility mainly stems from the possibility offered by our framework to express arbitrary constraints over variables. In this section, we illustrate these possibilities by describing how reordering constraints can easily be considered. As a first example, the Moses decoder uses a distortion limit to constrain the set of possible reorderings. This constraint “enforces (...) that the last word of a phrase chosen for translation cannot be more than d9 words from the leftmost untranslated word in the source” (Lopez, 2009) and is expressed as: ∀aijkl , ai0j0k0l0 s.t. k > k0, aijkl · ai0j0k0l0 · |j − i0 + 1| ≤ d, The maximum distortion limit strategy (Lopez, 2009) is also easily expressed and take the following form (assuming this constraint is parameterized by d): ∀l < m − 1, ai,j,k,l·ai0,j0,l+1,l0 · |i0 − j − 1| 71is%t e6hs.a distortion greater that Moses default distortion limit. alignment decisions enabled by the use of larger training corpora and phrase table. To evaluate the impact ofthe second heuristic, we computed the number of phrases discarded by Moses (be- cause of the default ttl limit) but used in the oracle hypotheses. In the English to French NEWSCO setting, they account for 34.11% of the total number of phrases used in the oracle hypotheses. When the oracle decoder is constrained to use the same phrase table as Moses, its BLEU-4 score drops to 42.78. This shows that filtering the phrase table prior to decoding discards many useful phrase pairs and is seriously limiting the best achievable performance, a conclusion shared with (Auli et al., 2009). Search Errors Search errors can be identified by comparing the score of the best hypothesis found by Moses and the score of the oracle hypothesis. If the score of the oracle hypothesis is higher, then there has been a search error; on the contrary, there has been an estimation error when the score of the oracle hypothesis is lower than the score of the best hypothesis found by Moses. 940 Based on the comparison of the score of Moses hypotheses and of oracle hypotheses for the English to French NEWSCO setting, our preliminary conclusion is that the number of search errors is quite limited: only about 5% of the hypotheses of our oracle decoder are actually getting a better score than Moses solutions. Again, this shows that the scoring function (model error) is one of the main bottleneck of current PBTS. Comparing these hypotheses is nonetheless quite revealing: while Moses mostly selects phrase pairs with high translation scores and generates monotonous alignments, our ILP decoder uses larger reorderings and less probable phrases to achieve better solutions: on average, the reordering score of oracle solutions is −5.74, compared to −76.78 fscoro rMeo osfe osr outputs. iGonivsen is −the5 weight assigned through MERT training to the distortion score, no wonder that these hypotheses are severely penalized. The Impact of Phrase Length The observed outputs do not only depend on decisions made during the search, but also on decisions made during training. One such decision is the specification of maximal length for the source and target phrases. In our framework, evaluating the impact of this decision is simple: it suffices to change the definition of indicator variables so as to consider only alignments between phrases of a given length. In the English-French NEWSCO setting, the most restrictive choice, when only alignments between single words are authorized, yields an oracle BLEU-4 of 48.68; however, authorizing phrases up to length 2 allows to achieve an oracle value of 66.57, very close to the score achieved when considering all extracted phrases (67.77). This is corroborated with a further analysis of our oracle alignments, which use phrases whose average source length is 1.21 words (respectively 1.31 for target words). If many studies have already acknowledged the predomi- nance of “small” phrases in actual translations, our oracle scores suggest that, for this language pair, increasing the phrase length limit beyond 2 or 3 might be a waste of computational resources. 4 Related Work To the best of our knowledge, there are only a few works that try to study the expressive power ofphrase-based machine translation systems or to provide tools for analyzing potential causes of failure. The approach described in (Auli et al., 2009) is very similar to ours: in this study, the authors propose to find and analyze the limits of machine translation systems by studying the reference reachability. A reference is reachable for a given system if it can be exactly generated by this system. Reference reachability is assessed using Moses in forced decoding mode: during search, all hypotheses that deviate from the reference are simply discarded. Even though the main goal of this study was to compare the search space of phrase-based and hierarchical systems, it also provides some insights on the impact of various search parameters in Moses, delivering conclusions that are consistent with our main results. As described in Section 1.2, these authors also propose a typology of the errors of a statistical translation systems, but do not attempt to provide methods for identifying them. The authors of (Turchi et al., 2008) study the learn- ing capabilities of Moses by extensively analyzing learning curves representing the translation performances as a function of the number of examples, and by corrupting the model parameters. Even though their focus is more on assessing the scoring function, they reach conclusions similar to ours: the current bottleneck of translation performances is not the representation power of the PBTS but rather in their scoring functions. Oracle decoding is useful to compute reachable pseudo-references in the context of discriminative training. This is the main motivation of (Tillmann and Zhang, 2006), where the authors compute high BLEU hypotheses by running a conventional decoder so as to maximize a per-sentence approximation of BLEU-4, under a simple (local) reordering model. Oracle decoding has also been used to assess the limitations induced by various reordering constraints in (Dreyer et al., 2007). To this end, the authors propose to use a beam-search based oracle decoder, which computes lower bounds of the best achievable BLEU-4 using dynamic programming techniques over finite-state (for so-called local and IBM constraints) or hierarchically structured (for ITG constraints) sets of hypotheses. Even 941 though the numbers reported in this study are not directly comparable with ours17, it seems that our decoder is not only conceptually much simpler, but also achieves much more optimistic lower-bounds of the oracle BLEU score. The approach described in (Li and Khudanpur, 2009) employs a similar technique, which is to guide a heuristic search in an hypergraph representing possible translation hypotheses with n-gram counts matches, which amounts to decoding with a n-gram model trained on the sole reference translation. Additional tricks are presented in this article to speed-up decoding. Computing oracle BLEU scores is also the subject of (Zens and Ney, 2005; Leusch et al., 2008), yet with a different emphasis. These studies are concerned with finding the best hypotheses in a word graph or in a consensus network, a problem that has various implications for multi-pass decoding and/or system combination techniques. The former reference describes an exponential approximate algorithm, while the latter proves the NPcompleteness of this problem and discuss various heuristic approaches. Our problem is somewhat more complex and using their techniques would require us to built word graphs containing all the translations induced by arbitrary segmentations and permutations of the source sentence. 5 Conclusions In this paper, we have presented a methodology for analyzing the errors of PBTS, based on the computation of an approximation of the BLEU-4 oracle score. We have shown that this approximation could be computed fairly accurately and efficiently using Integer Linear Programming techniques. Our main result is a confirmation of the fact that extant PBTS systems are expressive enough to achieve very high translation performance with respect to conventional quality measurements. The main efforts should therefore strive to improve on the way phrases and hypotheses are scored during training. This gives further support to attempts aimed at designing context-dependent scoring functions as in (Stroppa et al., 2007; Gimpel and Smith, 2008), or at attempts to perform discriminative training of feature-rich models. (Bangalore et al., 2007). We have shown that the examination of difficult-totranslate sentences was an effective way to detect errors or inconsistencies in the reference translations, making our approach a potential aid for controlling the quality or assessing the difficulty of test data. Our experiments have also highlighted the impact of various parameters. Various extensions of the baseline ILP program have been suggested and/or evaluated. In particular, the ILP formalism lends itself well to expressing various constraints that are typically used in conventional PBTS. In 17The best BLEU-4 oracle they achieve on Europarl German to English is approximately 48; but they considered a smaller version of the training corpus and the WMT’06 test set. our future work, we aim at using this ILP framework to systematically assess various search configurations. We plan to explore how replacing non-reachable references with high-score pseudo-references can improve discrim- inative training of PBTS. We are also concerned by determining how tight is our approximation of the BLEU4 score is: to this end, we intend to compute the best BLEU-4 score within the n-best solutions of the oracle decoding problem. Acknowledgments Warm thanks to Houda Bouamor for helping us with the annotation tool. This work has been partly financed by OSEO, the French State Agency for Innovation, under the Quaero program. References Tobias Achterberg. 2007. Constraint Integer Programming. Ph.D. thesis, Technische Universit a¨t Berlin. http : / / opus .kobv .de /tuberl in/vol ltexte / 2 0 0 7 / 16 11/ . Abhishek Arun and Philipp Koehn. 2007. Online learning methods for discriminative training of phrase based statistical machine translation. In Proc. of MT Summit XI, Copenhagen, Denmark. Michael Auli, Adam Lopez, Hieu Hoang, and Philipp Koehn. 2009. A systematic analysis of translation model search spaces. In Proc. of WMT, pages 224–232, Athens, Greece. Satanjeev Banerjee and Alon Lavie. 2005. METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proc. of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72, Ann Arbor, Michigan. Srinivas Bangalore, Patrick Haffner, and Stephan Kanthak. 2007. Statistical machine translation through global lexical selection and sentence reconstruction. In Proc. of ACL, pages 152–159, Prague, Czech Republic. L e´on Bottou and Olivier Bousquet. 2008. The tradeoffs oflarge scale learning. In Proc. of NIPS, pages 161–168, Vancouver, B.C., Canada. Chris Callison-Burch, Philipp Koehn, Christof Monz, and Josh Schroeder. 2009. Findings of the 2009 Workshop on Statistical Machine Translation. In Proc. of WMT, pages 1–28, Athens, Greece. David Chiang, Steve DeNeefe, Yee Seng Chan, and Hwee Tou Ng. 2008. Decomposability of translation metrics for improved evaluation and efficient algorithms. In Proc. of ECML, pages 610–619, Honolulu, Hawaii. John De Nero and Dan Klein. 2008. The complexity of phrase alignment problems. In Proc. of ACL: HLT, Short Papers, pages 25–28, Columbus, Ohio. Markus Dreyer, Keith B. Hall, and Sanjeev P. Khudanpur. 2007. Comparing reordering constraints for smt using efficient bleu oracle computation. In NAACL-HLT/AMTA Workshop on Syntax and Structure in Statistical Translation, pages 103– 110, Rochester, New York. 942 Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2001 . Fast decoding and optimal decoding for machine translation. In Proc. of ACL, pages 228–235, Toulouse, France. 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 Proc. of NAACL, pages 1–8, Edmonton, Canada. Kevin Gimpel and Noah A. Smith. 2008. Rich source-side context for statistical machine translation. In Proc. of WMT, pages 9–17, Columbus, Ohio. Philipp Koehn, Franz Josef Och, and Daniel Marcu. 2003. Statistical phrase-based translation. In Proc. of NAACL, pages 48–54, Edmonton, Canada. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris CallisonBurch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proc. of ACL, demonstration session. Philipp Koehn. 2004. Pharaoh: A beam search decoder for phrase-based statistical machine translation models. In Proc. of AMTA, pages 115–124, Washington DC. Shankar Kumar and William Byrne. 2005. Local phrase reordering models for statistical machine translation. In Proc. of HLT, pages 161–168, Vancouver, Canada. Alon Lavie, Kenji Sagae, and Shyamsundar Jayaraman. The significance of recall in automatic metrics for MT evaluation. In In Proc. of AMTA, pages 134–143, Washington DC. Gregor Leusch, Evgeny Matusov, and Hermann Ney. 2008. Complexity of finding the BLEU-optimal hypothesis in a confusion network. In Proc. of EMNLP, pages 839–847, Honolulu, Hawaii. Zhifei Li and Sanjeev Khudanpur. 2009. Efficient extraction of oracle-best translations from hypergraphs. In Proc. of NAACL, pages 9–12, Boulder, Colorado. Percy Liang, Alexandre Bouchard-C oˆt´ e, Dan Klein, and Ben Taskar. 2006. An end-to-end discriminative approach to machine translation. In Proc. of ACL, pages 761–768, Sydney, Australia. Adam Lopez. 2009. Translation as weighted deduction. In Proc. of EACL, pages 532–540, Athens, Greece. Franz Josef Och and Hermann Ney. 2003. A systematic comparison of various statistical alignment models. Comput. Linguist. , 29(1): 19–5 1. Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proc. of ACL, pages 160–167, Sapporo, Japan. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-jing Zhu. 2002. Bleu: A method for automatic evaluation of machine translation. Technical report, Philadelphia, Pennsylvania. D. Roth and W. Yih. 2005. Integer linear programming inference for conditional random fields. In Proc. of ICML, pages 737–744, Bonn, Germany. Nicolas Stroppa, Antal van den Bosch, and Andy Way. 2007. Exploiting source similarity for smt using context-informed features. In Andy Way and Barbara Proc. of TMI, pages Christoph Tillmann 231–240, Sk¨ ovde, and Tong Zhang. Gawronska, editors, Sweden. 2006. A discriminative global training algorithm for statistical mt. In Proc. of ACL, 721–728, Sydney, Australia. Turchi, Tijl De Bie, and Nello pages Marco Cristianini. 2008. Learn- ing performance of a machine translation system: a statistical and computational analysis. In Proc. of WMT, pages Columbus, Ohio. 35–43, Richard Zens and Hermann Ney. 2005. Word graphs for statistical machine translation. In Proc. of the ACL Workshop on Building and Using Parallel Texts, pages 191–198, Ann Arbor, Michigan. 943
5 0.17534339 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
Author: Lei Shi ; Rada Mihalcea ; Mingjun Tian
Abstract: In this paper, we introduce a method that automatically builds text classifiers in a new language by training on already labeled data in another language. Our method transfers the classification knowledge across languages by translating the model features and by using an Expectation Maximization (EM) algorithm that naturally takes into account the ambiguity associated with the translation of a word. We further exploit the readily available unlabeled data in the target language via semisupervised learning, and adapt the translated model to better fit the data distribution of the target language.
6 0.16825658 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
7 0.16024232 47 emnlp-2010-Example-Based Paraphrasing for Improved Phrase-Based Statistical Machine Translation
8 0.14424418 5 emnlp-2010-A Hybrid Morpheme-Word Representation for Machine Translation of Morphologically Rich Languages
9 0.14225361 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
10 0.13572647 22 emnlp-2010-Automatic Evaluation of Translation Quality for Distant Language Pairs
11 0.13189298 50 emnlp-2010-Facilitating Translation Using Source Language Paraphrase Lattices
12 0.12960362 99 emnlp-2010-Statistical Machine Translation with a Factorized Grammar
13 0.12465774 13 emnlp-2010-A Simple Domain-Independent Probabilistic Approach to Generation
14 0.1197084 55 emnlp-2010-Handling Noisy Queries in Cross Language FAQ Retrieval
15 0.1143134 29 emnlp-2010-Combining Unsupervised and Supervised Alignments for MT: An Empirical Study
16 0.11245698 17 emnlp-2010-An Efficient Algorithm for Unsupervised Word Segmentation with Branching Entropy and MDL
18 0.10861707 62 emnlp-2010-Improving Mention Detection Robustness to Noisy Input
19 0.10810387 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
20 0.1059875 53 emnlp-2010-Fusing Eye Gaze with Speech Recognition Hypotheses to Resolve Exophoric References in Situated Dialogue
topicId topicWeight
[(10, 0.023), (12, 0.051), (29, 0.051), (30, 0.029), (32, 0.01), (52, 0.025), (56, 0.097), (66, 0.088), (70, 0.387), (72, 0.055), (76, 0.036), (83, 0.012), (87, 0.025), (89, 0.013)]
simIndex simValue paperId paperTitle
1 0.87804347 14 emnlp-2010-A Tree Kernel-Based Unified Framework for Chinese Zero Anaphora Resolution
Author: Fang Kong ; Guodong Zhou
Abstract: This paper proposes a unified framework for zero anaphora resolution, which can be divided into three sub-tasks: zero anaphor detection, anaphoricity determination and antecedent identification. In particular, all the three sub-tasks are addressed using tree kernel-based methods with appropriate syntactic parse tree structures. Experimental results on a Chinese zero anaphora corpus show that the proposed tree kernel-based methods significantly outperform the feature-based ones. This indicates the critical role of the structural information in zero anaphora resolution and the necessity of tree kernel-based methods in modeling such structural information. To our best knowledge, this is the first systematic work dealing with all the three sub-tasks in Chinese zero anaphora resolution via a unified framework. Moreover, we release a Chinese zero anaphora corpus of 100 documents, which adds a layer of annotation to the manu- ally-parsed sentences in the Chinese Treebank (CTB) 6.0.
same-paper 2 0.74068844 19 emnlp-2010-Automatic Analysis of Rhythmic Poetry with Applications to Generation and Translation
Author: Erica Greene ; Tugba Bodrumlu ; Kevin Knight
Abstract: Tugba Bodrumlu Dept. of Computer Science Univ. of Southern California Los Angeles, CA 90089 bodrumlu@cs . usc . edu Kevin Knight Information Sciences Institute Univ. of Southern California 4676 Admiralty Way Marina del Rey, CA 90292 kn i @ i i ght s .edu from existing online poetry corpora. We use these patterns to generate new poems and translate exist- We employ statistical methods to analyze, generate, and translate rhythmic poetry. We first apply unsupervised learning to reveal word-stress patterns in a corpus of raw poetry. We then use these word-stress patterns, in addition to rhyme and discourse models, to generate English love poetry. Finally, we translate Italian poetry into English, choosing target realizations that conform to desired rhythmic patterns.
3 0.36169532 102 emnlp-2010-Summarizing Contrastive Viewpoints in Opinionated Text
Author: Michael Paul ; ChengXiang Zhai ; Roxana Girju
Abstract: This paper presents a two-stage approach to summarizing multiple contrastive viewpoints in opinionated text. In the first stage, we use an unsupervised probabilistic approach to model and extract multiple viewpoints in text. We experiment with a variety of lexical and syntactic features, yielding significant performance gains over bag-of-words feature sets. In the second stage, we introduce Comparative LexRank, a novel random walk formulation to score sentences and pairs of sentences from opposite viewpoints based on both their representativeness of the collection as well as their contrastiveness with each other. Exper- imental results show that the proposed approach can generate informative summaries of viewpoints in opinionated text.
4 0.36034077 120 emnlp-2010-What's with the Attitude? Identifying Sentences with Attitude in Online Discussions
Author: Ahmed Hassan ; Vahed Qazvinian ; Dragomir Radev
Abstract: Mining sentiment from user generated content is a very important task in Natural Language Processing. An example of such content is threaded discussions which act as a very important tool for communication and collaboration in the Web. Threaded discussions include e-mails, e-mail lists, bulletin boards, newsgroups, and Internet forums. Most of the work on sentiment analysis has been centered around finding the sentiment toward products or topics. In this work, we present a method to identify the attitude of participants in an online discussion toward one another. This would enable us to build a signed network representation of participant interaction where every edge has a sign that indicates whether the interaction is positive or negative. This is different from most of the research on social networks that has focused almost exclusively on positive links. The method is exper- imentally tested using a manually labeled set of discussion posts. The results show that the proposed method is capable of identifying attitudinal sentences, and their signs, with high accuracy and that it outperforms several other baselines.
5 0.35836998 49 emnlp-2010-Extracting Opinion Targets in a Single and Cross-Domain Setting with Conditional Random Fields
Author: Niklas Jakob ; Iryna Gurevych
Abstract: In this paper, we focus on the opinion target extraction as part of the opinion mining task. We model the problem as an information extraction task, which we address based on Conditional Random Fields (CRF). As a baseline we employ the supervised algorithm by Zhuang et al. (2006), which represents the state-of-the-art on the employed data. We evaluate the algorithms comprehensively on datasets from four different domains annotated with individual opinion target instances on a sentence level. Furthermore, we investigate the performance of our CRF-based approach and the baseline in a single- and cross-domain opinion target extraction setting. Our CRF-based approach improves the performance by 0.077, 0.126, 0.071 and 0. 178 regarding F-Measure in the single-domain extraction in the four domains. In the crossdomain setting our approach improves the performance by 0.409, 0.242, 0.294 and 0.343 regarding F-Measure over the baseline.
6 0.35814071 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
7 0.35729817 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
8 0.35680693 82 emnlp-2010-Multi-Document Summarization Using A* Search and Discriminative Learning
9 0.35666597 58 emnlp-2010-Holistic Sentiment Analysis Across Languages: Multilingual Supervised Latent Dirichlet Allocation
10 0.35461441 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
11 0.3494187 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
12 0.34811854 1 emnlp-2010-"Poetic" Statistical Machine Translation: Rhyme and Meter
13 0.34728593 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
14 0.34561142 25 emnlp-2010-Better Punctuation Prediction with Dynamic Conditional Random Fields
15 0.34547189 64 emnlp-2010-Incorporating Content Structure into Text Analysis Applications
16 0.34401536 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
17 0.3431631 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
18 0.34298804 20 emnlp-2010-Automatic Detection and Classification of Social Events
19 0.34257254 83 emnlp-2010-Multi-Level Structured Models for Document-Level Sentiment Classification
20 0.34180835 51 emnlp-2010-Function-Based Question Classification for General QA