emnlp emnlp2010 emnlp2010-13 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Gabor Angeli ; Percy Liang ; Dan Klein
Abstract: Percy Liang UC Berkeley Berkeley, CA 94720 pliang@cs.berkeley.edu Dan Klein UC Berkeley Berkeley, CA 94720 klein@cs.berkeley.edu We operate in a setting in which we are only given examples consisting of (i) a set of database records We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. We deployed our system in three different domains—Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-ofthe-art domain-specific systems both in terms of BLEU scores and human evaluation.
Reference: text
sentIndex sentText sentNum sentScore
1 edu We operate in a setting in which we are only given examples consisting of (i) a set of database records We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. [sent-6, score-0.944]
2 In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. [sent-7, score-0.23]
3 We deployed our system in three different domains—Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-ofthe-art domain-specific systems both in terms of BLEU scores and human evaluation. [sent-8, score-0.412]
4 1 Introduction In this paper, we focus on the problem of generating descriptive text given a world state represented by a set of database records. [sent-9, score-0.261]
5 While existing generation systems can be engineered to obtain good performance on particular domains (e. [sent-10, score-0.178]
6 (2005), inter alia), it is often difficult to adapt them across different domains. [sent-15, score-0.045]
7 Furthermore, content selection (what to say: see Barzilay and Lee (2004), Foster and White (2004), inter alia) and surface realization (how to say it: see Ratnaparkhi (2002), Wong and Mooney (2007), Chen and Mooney (2008), Lu et al. [sent-16, score-0.467]
8 Our goal is to build a simple, flexible system which is domain-independent and performs content selection and surface realization in a unified framework. [sent-19, score-0.422]
9 502 (input) and (ii) example human-generated text describing some of those records (output). [sent-20, score-0.333]
10 (2009) to automatically induce the correspondences between words in the text and the actual database records mentioned. [sent-22, score-0.386]
11 We break up the full generation process into a sequence of local decisions, training a log-linear classifier for each type of decision. [sent-23, score-0.22]
12 We use a simple but expressive set of domain-independent features, where each decision is allowed to depend on the entire history of previous decisions, as in the model of Ratnaparkhi (2002). [sent-24, score-0.084]
13 The first type chooses records from the database (macro content selection)—for example, wind speed, in the case of generating weather forecasts. [sent-27, score-0.848]
14 The second type chooses a subset of fields from a record (micro content selection)—e. [sent-28, score-0.69]
15 The third type chooses a suitable template to render the content (surface realization)— e. [sent-31, score-0.351]
16 , winds between [min] and [max] mph; templates are automatically extracted from training data. [sent-33, score-0.11]
17 We tested our approach in three domains: ROBOCUP, for sportscasting (Chen and Mooney, 2008); SUMTIME, for technical weather forecast generation (Reiter et al. [sent-34, score-0.534]
18 , 2005); and WEATHERGOV, for common weather forecast generation (Liang et al. [sent-35, score-0.455]
19 Our generation task is to map a world state s (input) to a text w (output). [sent-43, score-0.344]
20 Note that this mapping involves both content selection and surface realization. [sent-44, score-0.282]
21 5 on the combined task of content selection and generation, which is more than a two-fold improvement over a model similar to that of Liang et al. [sent-46, score-0.182]
22 2 Setup and Domains Our goal is to generate a text given a world state. [sent-50, score-0.148]
23 The world state, denoted s, is represented by a set of database records. [sent-51, score-0.201]
24 Define T to be a set of record types, wbahseere r ecaocrhd sr. [sent-52, score-0.343]
25 ec Doerdfi type t o ∈ eT a sise tas ofso rceicaoterdd twypitehs a wsehte roef feiaecldhs FIELDS(t). [sent-53, score-0.027]
26 tE ∈ach T r eisco asrdso r ∈ s has a record type r. [sent-54, score-0.37]
27 vd[ w, oisr represented by a sequence of tokenized words. [sent-58, score-0.034]
28 We use the term scenario to denote a world state s paired with a text w. [sent-59, score-0.278]
29 Example scenarios for each domain are de- tailed in Figure 1. [sent-61, score-0.081]
30 1 ROBOCUP: Sportscasting A world state in the ROBOCUP domain is a set of event records (meaning representations in the terminology of Chen and Mooney (2008)) generated by a robot soccer simulator. [sent-63, score-0.564]
31 For example, the record pass(arg1=pink1,arg2=pink5) denotes a passing event; records of this type (pass) have two fields: arg1 (the agent) and arg2 (the recipient). [sent-64, score-0.728]
32 As the game progresses, human commentators talk about some of the events in the game, e. [sent-65, score-0.09]
33 , purple3 made a bad pass that was picked off by pink9. [sent-67, score-0.029]
34 We used the dataset created by Chen and Mooney (2008), which contains 1919 scenarios from the 2001–2004 Robocup finals. [sent-68, score-0.086]
35 Each scenario consists of a single sentence representing a fragment of a commentary on the game, paired with a set of candidate records, which were recorded within five seconds of the commentary. [sent-69, score-0.093]
36 The records in the ROBOCUP dataset data were aligned by Chen and Mooney (2008). [sent-70, score-0.361]
37 en Ste ese Fleigcutrioen 1 (ina )t fhoirs domain is choosing the single record to talk about, and surface realization is talking about it. [sent-77, score-0.649]
38 (2005) developed a generation system and created the SUMTIME-METEO corpus, which consists of marine wind weather forecasts used by offshore oil rigs, generated by the output of weather simulators. [sent-80, score-0.793]
39 More specifically, these forecasts describe various aspects of the wind at different times during the forecast period. [sent-81, score-0.358]
40 The dataset consists of 469 scenarios, each containing on average |s| = 2o. [sent-83, score-0.028]
41 This task requires no content selection, only surface realization: The records are given in some fixed order and the task is to generate from each of these records in turn; of course, due to contextual dependencies, these records cannot be generated independently. [sent-87, score-1.203]
42 3 WEATHERGOV: Common Weather Forecasts In the WEATHERGOV domain, the world state contains detailed information about a local weather forecast (e. [sent-89, score-0.527]
43 The text is a short forecast report based on this information. [sent-93, score-0.113]
44 The world state is summarized by records which aggregate measurements over selected time intervals. [sent-96, score-0.541]
45 The dataset consists of 29,528 scenarios, each containing on average |s| = 36 records and 28. [sent-97, score-0.361]
46 While SUMTIME and WEATHERGOV are both weather domains, there are significant differences between the two. [sent-101, score-0.206]
47 SUMTIME forecasts are intended to be read by trained meteorologists, and thus the text is quite abbreviated. [sent-102, score-0.209]
48 On the other hand, WEATHERGOV texts are intended to be read by the general public and thus is more English-like. [sent-103, score-0.025]
49 Furthermore, SUMTIME does not require content selection, whereas content selection is a major focus of WEATHERGOV. [sent-104, score-0.286]
50 Indeed, on average, only 5 of 36 records are actually mentioned in a WEATHERGOV scenario. [sent-105, score-0.333]
51 Also, WEATHERGOV is more complex: The text is more varied, there are multiple record types, and there are about ten times as many records in each world state. [sent-106, score-0.824]
52 3 The Generation Process To model the process of generating a text w from a world state s, we decompose the generation process into a sequence of local decisions. [sent-112, score-0.455]
53 There are two aspects of this decomposition that we need to specify: (i) how the decisions are structured; and (ii) what pieces of information govern the decisions. [sent-113, score-0.326]
54 Figure 2 shows the pseudocode for the generation process, while Figure 3 depicts an example of the generation process on a WEATHERGOV scenario. [sent-115, score-0.34]
55 Each of these decisions is governed by a set of feature templates (see Figure 4), which are represented as functions of the current decision and past decisions. [sent-116, score-0.384]
56 These features can, in general, depend on the current decision and all previous decisions. [sent-120, score-0.084]
57 For example, referring to Figure 4, R2 features on the record choice depend on all the previous record decisions, and R5 features depend on the most recent template decision. [sent-121, score-0.934]
58 This is in contrast with most systems for content selection (Barzilay and Lee, 2004) and surface realization (Belz, 2008), where decisions must decompose locally according to either a graph or tree. [sent-122, score-0.689]
59 stkemyCpeovraetr1ur:se1ky:Ctemoveper (atimtuer=e(5tpimm-e6=a5mpm,m-6oadme,=m5i0n-=7454),mean=49,max=60) r1= skyCover1 r2= temperature1 r3= stop DecisionsField set F1= {mode} F2= {time,min} Template T1= hmostly cloudy ,i T2= hwith a low around [min] . [sent-126, score-0.236]
60 i Record mostly cloudy , r2= temperature1 F2= {time,min} Text with a low around 45 . [sent-127, score-0.103]
61 Specific active (nonzero) features for highlighted decisions (R1) (R2) (R3) (R4) (F1) (F2) (F2) (W1) Jr2 . [sent-128, score-0.262]
62 t = = = = = = = = temperature temperature temperature temperature temperature temperature temperature temperature and (r1. [sent-136, score-1.624]
63 vvv[[ mmmieanaxn]] ==] = =l m loewdoKiwKumK JJFr2 = {time, min}K JJFF2 == {{ttiimmee,, mmiinn}} and JJJFFF2 == {{ttiimmee,, mmiinn}} aanndd . [sent-154, score-0.041]
64 vv[[tmimine] = = l 5opwmKr2 JFBase(T2) = hwith a low. [sent-157, score-0.105]
65 a[mroiunn]d = [m loinw]K KiK JJCBoaaser(sTe()T=2 ) =w thwhi tah l oaw w[ti amroeu] adro [umnidn] i[Kmin]iK T2= hwith a low around [min]i (W2) J C Bo a saer(s Te 2 ()T = 2) h = w iht h w ita th l aoaow [ tia rm o e u ]n a d ro [ m u n in d ]i[ m a in nd ]i rK2 a. [sent-158, score-0.182]
66 ev][t=im5ep]m=-65apm K-6amK JJJCCBaoosaaerr(ssTee((TT2) = hwith a [time] (W3) lJJoCBgaopaslmer(s(Twe(itTh | cloudy ,) adro [umnidn] [min]i a. [sent-160, score-0.237]
67 vv][ =mtim lineo] w=K = l 5opwmK- Figure 3: The generation process on an example WEATHERGOV scenario. [sent-163, score-0.159]
68 The figure is divided into two parts: The upper part of the figure shows the generation of text from the world state via a sequence of seven decisions (in boxes). [sent-164, score-0.614]
69 Three of these decisions are highlighted and the features that govern these decisions are shown in the lower part of the figure. [sent-165, score-0.588]
70 Note that different decisions in the generation process would result in different features being active (nonzero). [sent-166, score-0.395]
71 v]=[f ] =K ∗K for f ∈ Fi, h ∈ {Base, Coarse} JJlhhog(( TTplm) (=fir ∗stK wnodr dr vin[ fT]i = =| p∗Krevious two words) and template decisions. [sent-191, score-0.156]
72 t = ∗K, ∗ rdaensgcersip toiovenr, aanll d re fcoorrmda tly dpeessc rTip ). [sent-193, score-0.023]
73 t oNno otfat aio sne: oJefK reaettuurrness 1 o bift athinee dex bpyre rsasniognin eg i∗s torvueer apnods 0ib life i tv ailsu efasls (ef. [sent-194, score-0.081]
74 teiyo N nao orteaf tui aos sened:t Jtooef Kc f erreaettauutr en ssfe, a1 otub iftrae tshin eeade uxto bpmyre rastasniicoganilnly eg ∗iasc r torovuseesr dpaonomdss 0aibin ilefs. [sent-197, score-0.035]
75 from the world state s according to features of the following types: 3. [sent-205, score-0.208]
76 1 Record Decisions R1 captures Record decisions are responsible for macro content content selection. [sent-206, score-0.583]
77 Each record decision chooses a record ri windSpeed the discourse selection; for coherence example, tends to follow we windDir aspect learn of that (but not al- 505 ways). [sent-207, score-0.893]
78 R2 captures an unordered notion of coherence—simply which sets of record types are preferable; for example, we learn that rainChance is not generated if sleetChance already was mentioned. [sent-208, score-0.388]
79 R3 is a coarser version of R2, capturing how likely it is to propose a record of a type that has already been generated. [sent-209, score-0.37]
80 R4 captures the important aspect of content selection that the records chosen depend on their field values;1 for example, we learn that snowChance is not chosen unless there is snow. [sent-210, score-0.875]
81 R5 allows the language model to indicate whether a STOP record is appropriate; this helps prevent sentences from ending abruptly. [sent-211, score-0.343]
82 2 Field Set Decisions Field set decisions are responsible for micro content selection, i. [sent-213, score-0.434]
83 Each field set decision chooses a subset of fields Fi from the set of fields FIELDS(ri. [sent-216, score-0.611]
84 These decisions are made based on two types of features: F1 captures which sets of fields are talked about together; for example, we learn that {mean} aabndo {min, max} are preferred lfeiealdrn st hetast {fmore atnhe} windSpeed m reacxo}rd. [sent-218, score-0.433]
85 a By defining f feiaetludre sse on ftohre entire field set, we can capture any correlation structure over the fields; in contrast, Liang et al. [sent-219, score-0.205]
86 (2009) generates a sequence of fields in which a field can only depend on the previous one. [sent-220, score-0.437]
87 F2 allows the field set to be chosen based on the values of the fields, analogously to R4. [sent-221, score-0.237]
88 A template is a sequence of elements, where each ele- ment is either a word (e. [sent-224, score-0.19]
89 Given the record ri and field set Fi that we are generating from, the goal is to choose a template Ti (Section 4. [sent-229, score-0.778]
90 The features that govern the choice of Ti are as follows: W1 captures a priori preferences for generation templates given field sets. [sent-232, score-0.586]
91 1We map a numeric field value onto one of five categories (very-low, low, medium, high, or very-high) based on its value with respect to the mean and standard deviation of values of that field in the training data. [sent-234, score-0.41]
92 506 BASE(Ti) denotes the template Ti itself, thus allowing us to remember exactly which templates were useful. [sent-235, score-0.291]
93 To guard against overfitting, we also use COARSE(Ti), which maps Ti to a coarsened version of Ti, in which more words are replaced with their associated fields (see Figure 5 for an example). [sent-236, score-0.152]
94 W2 captures a dependence on the values of fields in the field set, and is analogous to R4 and F2. [sent-237, score-0.402]
95 After Ti has been chosen, each field in the template is replaced with a word given the corresponding field value in the world state. [sent-239, score-0.714]
96 In particular, a word is chosen from the parameters learned in the model of Liang et al. [sent-240, score-0.032]
97 In the example in Figure 3, the [min] field in T2 has value 44, which is rendered to the word 45 (rounding and other noisy deviations are common in the WEATHERGOV domain). [sent-242, score-0.251]
98 4 Learning a Probabilistic Model Having described all the features, we now present a conditional probabilistic model over texts w given world states s (Section 4. [sent-243, score-0.148]
99 1 Model Recall from Section 3 that the generation process generates r1, F1, T1, r2, F2 , T2, . [sent-249, score-0.159]
100 To unify notation, denote this sequence of decisions as d = (d1, . [sent-253, score-0.27]
wordName wordTfidf (topN-words)
[('record', 0.343), ('records', 0.333), ('weathergov', 0.316), ('decisions', 0.236), ('weather', 0.206), ('field', 0.205), ('temperature', 0.203), ('forecasts', 0.184), ('robocup', 0.184), ('sumtime', 0.184), ('template', 0.156), ('fields', 0.152), ('world', 0.148), ('realization', 0.14), ('generation', 0.136), ('forecast', 0.113), ('templates', 0.11), ('hwith', 0.105), ('content', 0.104), ('surface', 0.1), ('govern', 0.09), ('cloudy', 0.079), ('sportscasting', 0.079), ('selection', 0.078), ('berkeley', 0.076), ('ri', 0.074), ('mooney', 0.071), ('min', 0.07), ('reiter', 0.068), ('chooses', 0.064), ('ti', 0.064), ('liang', 0.063), ('wind', 0.061), ('macro', 0.061), ('micro', 0.061), ('state', 0.06), ('scenarios', 0.058), ('uc', 0.056), ('database', 0.053), ('adro', 0.053), ('jjjrrri', 0.053), ('mmiinn', 0.053), ('umnidn', 0.053), ('windspeed', 0.053), ('game', 0.047), ('depend', 0.046), ('captures', 0.045), ('pseudocode', 0.045), ('nonzero', 0.045), ('alia', 0.045), ('inter', 0.045), ('ratnaparkhi', 0.045), ('chen', 0.044), ('talk', 0.043), ('domains', 0.042), ('aanndd', 0.041), ('belz', 0.041), ('scenario', 0.04), ('coarse', 0.038), ('decision', 0.038), ('fi', 0.038), ('hierarchically', 0.037), ('ste', 0.037), ('eg', 0.035), ('rj', 0.035), ('sr', 0.035), ('sequence', 0.034), ('responsible', 0.033), ('chosen', 0.032), ('coherence', 0.031), ('decompose', 0.031), ('paired', 0.03), ('pass', 0.029), ('stop', 0.028), ('dataset', 0.028), ('barzilay', 0.027), ('type', 0.027), ('highlighted', 0.026), ('bleu', 0.026), ('ii', 0.026), ('tm', 0.025), ('denotes', 0.025), ('intended', 0.025), ('lm', 0.024), ('around', 0.024), ('domain', 0.023), ('process', 0.023), ('rain', 0.023), ('pliang', 0.023), ('sne', 0.023), ('rendered', 0.023), ('aanll', 0.023), ('athinee', 0.023), ('commentary', 0.023), ('deviations', 0.023), ('dex', 0.023), ('din', 0.023), ('eem', 0.023), ('fir', 0.023), ('iasc', 0.023)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999994 13 emnlp-2010-A Simple Domain-Independent Probabilistic Approach to Generation
Author: Gabor Angeli ; Percy Liang ; Dan Klein
Abstract: Percy Liang UC Berkeley Berkeley, CA 94720 pliang@cs.berkeley.edu Dan Klein UC Berkeley Berkeley, CA 94720 klein@cs.berkeley.edu We operate in a setting in which we are only given examples consisting of (i) a set of database records We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. We deployed our system in three different domains—Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-ofthe-art domain-specific systems both in terms of BLEU scores and human evaluation.
2 0.07922323 64 emnlp-2010-Incorporating Content Structure into Text Analysis Applications
Author: Christina Sauper ; Aria Haghighi ; Regina Barzilay
Abstract: In this paper, we investigate how modeling content structure can benefit text analysis applications such as extractive summarization and sentiment analysis. This follows the linguistic intuition that rich contextual information should be useful in these tasks. We present a framework which combines a supervised text analysis application with the induction of latent content structure. Both of these elements are learned jointly using the EM algorithm. The induced content structure is learned from a large unannotated corpus and biased by the underlying text analysis task. We demonstrate that exploiting content structure yields significant improvements over approaches that rely only on local context.1
3 0.07123521 99 emnlp-2010-Statistical Machine Translation with a Factorized Grammar
Author: Libin Shen ; Bing Zhang ; Spyros Matsoukas ; Jinxi Xu ; Ralph Weischedel
Abstract: In modern machine translation practice, a statistical phrasal or hierarchical translation system usually relies on a huge set of translation rules extracted from bi-lingual training data. This approach not only results in space and efficiency issues, but also suffers from the sparse data problem. In this paper, we propose to use factorized grammars, an idea widely accepted in the field of linguistic grammar construction, to generalize translation rules, so as to solve these two problems. We designed a method to take advantage of the XTAG English Grammar to facilitate the extraction of factorized rules. We experimented on various setups of low-resource language translation, and showed consistent significant improvement in BLEU over state-ofthe-art string-to-dependency baseline systems with 200K words of bi-lingual training data.
4 0.054834813 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
Author: Xian Qian ; Qi Zhang ; Yaqian Zhou ; Xuanjing Huang ; Lide Wu
Abstract: Many sequence labeling tasks in NLP require solving a cascade of segmentation and tagging subtasks, such as Chinese POS tagging, named entity recognition, and so on. Traditional pipeline approaches usually suffer from error propagation. Joint training/decoding in the cross-product state space could cause too many parameters and high inference complexity. In this paper, we present a novel method which integrates graph structures of two subtasks into one using virtual nodes, and performs joint training and decoding in the factorized state space. Experimental evaluations on CoNLL 2000 shallow parsing data set and Fourth SIGHAN Bakeoff CTB POS tagging data set demonstrate the superiority of our method over cross-product, pipeline and candidate reranking approaches.
5 0.05061616 52 emnlp-2010-Further Meta-Evaluation of Broad-Coverage Surface Realization
Author: Dominic Espinosa ; Rajakrishnan Rajkumar ; Michael White ; Shoshana Berleant
Abstract: We present the first evaluation of the utility of automatic evaluation metrics on surface realizations of Penn Treebank data. Using outputs of the OpenCCG and XLE realizers, along with ranked WordNet synonym substitutions, we collected a corpus of generated surface realizations. These outputs were then rated and post-edited by human annotators. We evaluated the realizations using seven automatic metrics, and analyzed correlations obtained between the human judgments and the automatic scores. In contrast to previous NLG meta-evaluations, we find that several of the metrics correlate moderately well with human judgments of both adequacy and fluency, with the TER family performing best overall. We also find that all of the metrics correctly predict more than half of the significant systemlevel differences, though none are correct in all cases. We conclude with a discussion ofthe implications for the utility of such metrics in evaluating generation in the presence of variation. A further result of our research is a corpus of post-edited realizations, which will be made available to the research community. 1 Introduction and Background In building surface-realization systems for natural language generation, there is a need for reliable automated metrics to evaluate the output. Unlike in parsing, where there is usually a single goldstandard parse for a sentence, in surface realization there are usually many grammatically-acceptable ways to express the same concept. This parallels the task of evaluating machine-translation (MT) systems: for a given segment in the source language, 564 there are usually several acceptable translations into the target language. As human evaluation of translation quality is time-consuming and expensive, a number of automated metrics have been developed to evaluate the quality of MT outputs. In this study, we investigate whether the metrics developed for MT evaluation tasks can be used to reliably evaluate the outputs of surface realizers, and which of these metrics are best suited to this task. A number of surface realizers have been developed using the Penn Treebank (PTB), and BLEU scores are often reported in the evaluations of these systems. But how useful is BLEU in this context? The original BLEU study (Papineni et al., 2001) scored MT outputs, which are of generally lower quality than grammar-based surface realizations. Furthermore, even for MT systems, the usefulness of BLEU has been called into question (Callison-Burch et al., 2006). BLEU is designed to work with multiple reference sentences, but in treebank realization, there is only a single reference sentence available for comparison. A few other studies have investigated the use of such metrics in evaluating the output of NLG systems, notably (Reiter and Belz, 2009) and (Stent et al., 2005). The former examined the performance of BLEU and ROUGE with computer-generated weather reports, finding a moderate correlation with human fluency judgments. The latter study applied several MT metrics to paraphrase data from Barzilay and Lee’s corpus-based system (Barzilay and Lee, 2003), and found moderate correlations with human adequacy judgments, but little correlation with fluency judgments. Cahill (2009) examined the performance of six MT metrics (including BLEU) in evaluating the output of a LFG-based surface realizer for ProceMedITin,g Ms oasfs thaceh 2u0se1t0ts C,o UnSfAer,e n9c-e1 on O Ectmobpeir ic 2a0l1 M0.e ?tc ho2d0s10 in A Nsastoucira tlio Lnan fogru Cagoem Ppruotcaetisosninagl, L pinag eusis 5t6ic4s–574, German, also finding only weak correlations with the human judgments. To study the usefulness of evaluation metrics such as BLEU on the output of grammar-based surface realizers used with the PTB, we assembled a corpus of surface realizations from three different realizers operating on Section 00 of the PTB. Two human judges evaluated the adequacy and fluency of each of the realizations with respect to the reference sentence. The realizations were then scored with a number of automated evaluation metrics developed for machine translation. In order to investigate the correlation of targeted metrics with human evaluations, and gather other acceptable realizations for future evaluations, the judges manually repaired each unacceptable realization during the rating task. In contrast to previous NLG meta-evaluations, we found that several of the metrics correlate moderately well with human judgments of both adequacy and fluency, with the TER family performing best. However, when looking at statistically significant system-level differences in human judgments, we found that some of the metrics get some of the rankings correct, but none get them all correct, with different metrics making different ranking errors. This suggests that multiple metrics should be routinely consulted when comparing realizer systems. Overall, our methodology is similar to that of previous MT meta-evaluations, in that we collected human judgments of system outputs, and compared these scores with those assigned by automatic metrics. A recent alternative approach to paraphrase evaluation is ParaMetric (Callison-Burch et al., 2008); however, it requires a corpus of annotated (aligned) paraphrases (which does not yet exist for PTB data), and is arguably focused more on paraphrase analysis than paraphrase generation. The plan of the paper is as follows: Section 2 discusses the preparation of the corpus of surface realizations. Section 3 describes the human evaluation task and the automated metrics applied. Sections 4 and 5 present and discuss the results of these evaluations. We conclude with some general observations about automatic evaluation of surface realizers, and some directions for further research. 565 2 Data Preparation We collected realizations of the sentences in Section 00 of the WSJ corpus from the following three sources: 1. OpenCCG, a CCG-based chart realizer (White, 2006) 2. The XLE Generator, a LFG-based system developed by Xerox PARC (Crouch et al., 2008) 3. WordNet synonym substitutions, to investigate how differences in lexical choice compare to grammar-based variation.1 Although all three systems used Section 00 of the PTB, they were applied with various parameters (e.g., language models, multiple-output versus single-output) and on different input structures. Accordingly, our study does not compare OpenCCG to XLE, or either of these to the WordNet system. 2.1 OpenCCG realizations OpenCCG is an open source parsing/realization library with multimodal extensions to CCG (Baldridge, 2002). The OpenCCG chart realizer takes logical forms as input and produces strings by combining signs for lexical items. Alternative realizations are scored using integrated n-gram and perceptron models. For robustness, fragments are greedily assembled when necessary. Realizations were generated from 1,895 gold standard logical forms, created by constrained parsing of development-section derivations. The following OpenCCG models (which differ essentially in the way the output is ranked) were used: 1. Baseline 1: Output ranked by a trigram word model 2. Baseline 2: Output ranked using three language models (3-gram words 3-gram words with named entity class replacement factored language model of words, POS tags and CCG supertags) + + 1Not strictly surface realizations, since they do not involve an abstract input specification, but for simplicity we refer to them as realizations throughout. 3. Baseline 3: Perceptron with syntax features and the three LMs mentioned above 4. Perceptron full-model: n-best realizations ranked using perceptron with syntax features and the three n-gram models, as well as discriminative n-grams The perceptron model was trained on sections 0221 of the CCGbank, while a grammar extracted from section 00-21 was used for realization. In addition, oracle supertags were inserted into the chart during realization. The purpose of such a non-blind testing strategy was to evaluate the quality of the output produced by the statistical ranking models in isolation, rather than focusing on grammar coverage, and avoid the problems associated with lexical smoothing, i.e. lexical categories in the development section not being present in the training section. To enrich the variation in the generated realizations, dative-alternation was enforced during realization by ensuring alternate lexical categories of the verb in question, as in the following example: (1) the executives gave [the chefs] [a standing ovation] (2) the executives gave [a standing ovation] [to the chefs] 2.2 XLE realizations The corpus of realizations generated by the XLE system contained 42,527 surface realizations of approximately 1,421 section 00 sentences (an average of 30 per sentence), initially unranked. The LFG f-structures used as input to the XLE generator were derived from automatic parses, as described in (Riezler et al., 2002). The realizations were first tokenized using Penn Treebank conventions, then ranked using perplexities calculated from the same trigram word model used with OpenCCG. For each sentence, the top 4 realizations were selected. The XLE generator provides an interesting point of comparison to OpenCCG as it uses a manuallydeveloped grammar with inputs that are less abstract but potentially noisier, as they are derived from automatic parses rather than gold-standard ones. 566 2.3 WordNet synonymizer To produce an additional source of variation, the nouns and verbs of the sentences in section 00 of the PTB were replaced with all of their WordNet synonyms. Verb forms were generated using verb stems, part-of-speech tags, and the morphg tool.2 These substituted outputs were then filtered using the n-gram data which Google Inc. has made available.3 Those without any 5-gram matches centered on the substituted word (or 3-gram matches, in the case of short sentences) were eliminated. 3 Evaluation From the data sources described in the previous sec- tion, a corpus of realizations to be evaluated by the human judges was constructed by randomly choosing 305 sentences from section 00, then selecting surface realizations of these sentences using the following algorithm: 1. Add OpenCCG’s best-scored realization. 2. Add other OpenCCG realizations until all four models are represented, to a maximum of 4. 3. Add up to 4 realizations from either the XLE system or the WordNet pool, chosen randomly. The intent was to give reasonable coverage of all realizer systems discussed in Section 2 without overloading the human judges. “System” here means any instantiation that emits surface realizations, including various configurations of OpenCCG (using different language models or ranking systems), and these can be multiple-output, such as an n-best list, or single-output (best-only, worst-only, etc.). Accordingly, more realizations were selected from the OpenCCG realizer because 5 different systems were being represented. Realizations were chosen randomly, rather than according to sentence types or other criteria, in order to produce a representative sample of the corpus. In total, 2,114 realizations were selected for evaluation. 2http : //www. informatics . sussex. ac .uk/ re search/ groups / nlp / carro l /morph .html l 3http : //www . ldc . upenn .edu/Catalog/docs/ LDC2 0 0 6T 13 / readme .txt 3.1 Human judgments Two human judges evaluated each surface realization on two criteria: adequacy, which represents the extent to which the output conveys all and only the meaning of the reference sentence; and fluency, the extent to which it is grammatically acceptable. The realizations were presented to the judges in sets containing a reference sentence and the 1-8 outputs selected for that sentence. To aid in the evaluation of adequacy, one sentence each of leading and trailing context were displayed. Judges used the guidelines given in Figure 1, based on the scales developed by the NIST Machine Translation Evaluation Workshop. In addition to rating each realization on the two five-point scales, each judge also repaired each output which he or she did not judge to be fully adequate and fluent. An example is shown in Figure 2. These repairs resulted in new reference sentences for a substantial number of sentences. These repaired realizations were later used to calculate targeted versions of the evaluation metrics, i.e., using the repaired sentence as the reference sentence. Although targeted metrics are not fully automatic, they are of interest because they allow the evaluation algorithm to focus on what is actually wrong with the input, rather than all textual differences. Notably, targeted TER (HTER) has been shown to be more consistent with human judgments than human annotators are with one another (Snover et al., 2006). 3.2 Automatic evaluation The realizations were also evaluated using seven automatic metrics: • IBM’s BLEU, which scores a hypothesis by counting n-gram matches with the reference sentence (Papineni et al., 2001), with smoothing as described in (Lin and Och, 2004) • • • • • • The NIST n-gram evaluation metric, similar to BLEU, but rewarding rarer n-gram matches, and using a different length penalty METEOR, which measures the harmonic mean of unigram precision and recall, with a higher weight for recall (Banerjee and Lavie, 2005) 567 TER (Translation Edit Rate), a measure of the number of edits required to transform a hypothesis sentence into the reference sentence (Snover et al., 2006) TERP, an augmented version of TER which performs phrasal substitutions, stemming, and checks for synonyms, among other improvements (Snover et al., 2009) TERPA, an instantiation of TERP with edit weights optimized for correlation with adequacy in MT evaluations GTM (General Text Matcher), a generaliza- tion of the F-measure that rewards contiguous matching spans (Turian et al., 2003) Additionally, targeted versions of BLEU, METEOR, TER, and GTM were computed by using the human-repaired outputs as the reference set. The human repair was different from the reference sentence in 193 cases (about 9% of the total), and we expected this to result in better scores and correlations with the human judgments overall. 4 Results 4.1 Human judgments Table 1 summarizes the dataset, as well as the mean adequacy and fluency scores garnered from the human evaluation. Overall adequacy and fluency judgments were high (4.16, 3.63) for the realizer systems on average, and the best-rated realizer systems achieved mean fluency scores above 4. 4.2 Inter-annotator agreement Inter-annotator agreement was measured using the κ-coefficient, which is commonly used to measure the extent to which annotators agree in category P(1A−)P−(PE()E), judgment tasks. κ is defined as where P(A) is the observed agreement 1 b−etPw(eEe)n annotators and P(E) is the probability of agreement due to chance (Carletta, 1996). Chance agreement for this data is calculated by the method discussed in Carletta’s squib. However, in previous work in MT meta-evaluation, Callison-Burch et al. (2007), assume the less strict criterion of uniform chance agreement, i.e. for a five-point scale. They also 51 Score Adequacy Fluency 5All the meaning of the referencePerfectly grammatical 4 Most of the meaning Awkward or non-native; punctuation errors 3 Much of the meaning Agreement errors or minor syntactic problems 2 Meaning substantially different Major syntactic problems, such as missing words 1 Meaning completely different Completely ungrammatical Figure Ref. Realiz. Repair 1: Rating scale and guidelines It wasn’t clear how NL and Mr. Simmons would respond if Georgia Gulf spurns them again It weren’t clear how NL and Mr. Simmons would respond if Georgia Gulf again spurns them It wasn’t clear how NL and Mr. Simmons would respond if Georgia Gulf again spurns them Figure 2: Example of repair introduce the notion of “relative” κ, which measures how often two or more judges agreed that A > B, A = B, or A < B for two outputs A and B, irrespective of the specific values given on the five-point scale; here, uniform chance agreement is taken to be We report both absolute and relative κ in Table 2, using actual chance agreement rather than uniform chance agreement. 31. The κ scores of0.60 for adequacy and 0.63 for fluency across the entire dataset represent “substantial” agreement, according to the guidelines discussed in (Landis and Koch, 1977), better than is typically reported for machine translation evaluation tasks; for example, Callison-Burch et al. (2007) reported “fair” agreement, with κ = 0.281 for fluency and κ = 0.307 for adequacy (relative). Assuming the uniform chance agreement that the previously cited work adopts, our inter-annotator agreements (both absolute and relative) are still higher. This is likely due to the generally high quality of the realizations evaluated, leading to easier judgments. 4.3 Correlation with automatic evaluation To determine how well the automatic evaluation methods described in Section 3 correlate with the human judgments, we averaged the human judgments for adequacy and fluency, respectively, for each of the rated realizations, and then computed both Pearson’s correlation coefficient and Spearman’s rank correlation coefficient between these scores and each of the metrics. Spearman’s correlation makes fewer assumptions about the distribu- tion of the data, but may not reflect a linear rela568 tionship that is actually present. Both are frequently reported in the literature. Due to space constraints, we show only Spearman’s correlation, although the TER family scored slightly better on Pearson’s coefficient, relatively. The results for Spearman’s correlation are given in Table 3. Additionally, the average scores for adequacy and fluency were themselves averaged into a single score, following (Snover et al., 2009), and the Spearman’s correlation of each of the automatic metrics with these scores are given in Table 4. All reported correlations are significant at p < 0.001. 4.4 Bootstrap sampling of correlations For each of the sub-corpora shown in Table 1, we computed confidence intervals for the correlations between adequacy and fluency human scores with selected automatic metrics (BLEU, HBLEU, TER, TERP, and HTER) as described in (Koenh, 2004). We sampled each sub-corpus 1000 times with replace- ment, and calculated correlations between the rankings induced by the human scores and those induced by the metrics for each reference sentence. We then used these coefficients to estimate the confidence interval, after excluding the top 25 and bottom 25 coefficients, following (Lin and Och, 2004). The results of this for the BLEU metric are shown in Table 5. We determined which correlations lay within the 95% confidence interval of the best performing metric in each row of Table Table 3; these figures are italicized. 5 Discussion 5.1 Human judgments of systems The results for the four OpenCCG perceptron models mostly confirm those reported in (White and Rajkumar, 2009), with one exception: the B-3 model was below B-2, though the P-B (perceptron-best) model still scored highest. This may have been due to differences in the testing scenario. None of the differences in adequacy scores among the individual systems are significant, with the exception of the WordNet system. In this case, the lack of wordsense disambiguation for the substituted words results in a poor overall adequacy score (e.g., wage floor → wage story). Conversely, it scores highest ffoloro fluency, as substituting a noun or tve srcbo rwesith h a synonym does not usually introduce ungrammaticality. 5.2 Correlations of human judgments with MT metrics Of the non-human-targeted metrics evaluated, BLEU and TER/TERP demonstrate the highest correlations with the human judgments of fluency (r = 0.62, 0.64). The TER family of evaluation metrics have been observed to perform very well in MTevaluation tasks, and although the data evaluated here differs from typical MT data in some important ways, the correlation of TERP with the human judgments is substantial. In contrast with previous MT evaluations where TERP performs considerably better than TER, these scored close to equal on our data, possibly because TERP’s stem, synonym, and paraphrase matching are less useful when most of the variation is syntactic. The correlations with BLEU and METEOR are lower than those reported in (Callison-Burch et al., 2007); in that study, BLEU achieved adequacy and fluency correlations of 0.690 and 0.722, respectively, and METEOR achieved 0.701 and 0.719. The correlations for these metrics might be expected to be lower for our data, since overall quality is higher, making the metrics’ task more difficult as the outputs involve subtler differences between acceptable and unacceptable variation. The human-targeted metrics (represented by the prefixed H in the data tables) correlated even more strongly with the human judgments, compared to the non-targeted versions. HTER demonstrated the best 569 correlation with realizer fluency (r = 0.75). For several kinds of acceptable variation involving the rearrangement of constituents (such as dative shift), TERP gives a more reasonable score than BLEU, due to its ability to directly evaluate phrasal shifts. The following realization was rated 4.5 for fluency, and was more correctly ranked by TERP than BLEU: (3) Ref: The deal also gave Mitsui access to a high-tech medical product. (4) Realiz.: The deal also gave access to a high-tech medical product to Mitsui. For each reference sentence, we compared the ranking of its realizations induced from the human scores to the ranking induced from the TERP score, and counted the rank errors by the latter, informally categorizing them by error type (see Table 7). In the 50 sentences with the highest numbers of rank errors, 17 were affected by punctuation differences, typically involving variation in comma placement. Human fluency judgments of outputs with only punctuation problems were generally high, and many realizations with commas inserted or removed were rated fully fluent by the annotators. However, TERP penalizes such insertions or deletions. Agreement errors are another frequent source of ranking errors for TERP. The human judges tended to harshly penalize sentences with number-agreement or tense errors, whereas TERP applies only a single substitution penalty for each such error. We expect that with suitable optimization of edit weights to avoid over-penalizing punctuation shifts and underpenalizing agreement errors, TERP would exhibit an even stronger correlation with human fluency judgments. None of the evaluation metrics can distinguish an acceptable movement of a word or constituent from an unacceptable movement, with only one reference sentence. A substantial source of error for both TERP and BLEU is variation in adverbial placement, as shown in (7). Similar errors are seen with prepositional phrases and some commonly-occurring temporal adverbs, which typically admit a number of variations in placement. Another important example of acceptable variation which these metrics do not generally rank correctly is dative alternation: Ref. We need to clarify what exactly is wrong with it. Realiz. Flu. TERP BLEU We need to clarify exactly what is wrong with it.50.10.5555 We need to clarify exactly what ’s wrong with it. 5 0.2 0.4046 (7) We need to clarify what , exactly , is wrong with it. 5 0.2 0.5452 We need to clarify what is wrong with it exactly. 4.5 0.1 0.6756 We need to clarify what exactly , is wrong with it. 4 0.1 0.7017 We need to clarify what , exactly is wrong with it. 4 0.1 0.7017 We needs to clarify exactly what is wrong with it. (5) Ref. When test booklets were passed out 48 hours ahead of time, she says she copied questions in the social studies section and gave the answers to students. (6) Realiz. When test booklets were passed out 48 hours ahead of time , she says she copied questions in the social studies section and gave students the answers. The correlations of each of the metrics with the human judgments of fluency for the realizer systems indicate at least a moderate relationship, in contrast with the results reported in (Stent et al., 2005) for paraphrase data, which found an inverse correlation for fluency, and (Cahill, 2009) for the output ofa surface realizer for German, which found only a weak correlation. However, the former study employed a corpus-based paraphrase generation system rather than grammar-driven surface realizers, and the resulting paraphrases exhibited much broader variation. In Cahill’s study, the outputs of the realizer were almost always grammatically correct, and the automated evaluation metrics were ranking markedness instead of grammatical acceptability. 5.3 System-level comparisons In order to investigate the efficacy of the metrics in ranking different realizer systems, or competing realizations from the same system generated using different ranking models, we considered seven different “systems” from the whole dataset of realizations. These consisted of five OpenCCG-based realizations (the best realization from three baseline models, and the best and the worst realization from the full perceptron model), and two XLE-based sys- tems (the best and the worst realization, after ranking the outputs of the XLE realizer with an n-gram model). The mean of the combined adequacy and 570 3 0.103 0.346 fluency scores of each of these seven systems was compared with that of every other system, resulting in 21 pairwise comparisons. Then Tukey’s HSD test was performed to determine the systems which differed significantly in terms of the average adequacy and fluency rating they received.4 The test revealed five pairwise comparisons where the scores were significantly different. Subsequently, for each of these systems, an overall system-level score for each of the MT metrics was calculated. For the five pairwise comparisons where the adequacy-fluency group means differed significantly, we checked whether the metric ranked the systems correctly. Table 8 shows the results of a pairwise comparison between the ranking induced by each evaluation metric, and the ranking induced by the human judgments. Five of the seven non- targeted metrics correctly rank more than half of the systems. NIST, METEOR, and GTM get the most comparisons right, but neither NIST nor GTM correctly rank the OpenCCG-baseline model 1 with respect to the XLE-best model. TER and TERP get two of the five comparisons correct, and they incorrectly rank two of the five OpenCCG model comparisons, as well as the comparison between the XLE-worst and OpenCCG-best systems. For the targeted metrics, HNIST is correct for all five comparisons, while neither HBLEU nor HMETEOR correctly rank all the OpenCCG models. On the other hand, HTER and HGTM incorrectly rank the XLE-best system versus OpenCCG-based models. In summary, some of the metrics get some of the rankings correct, but none of the non-targeted metrics get all of them correct. Moreover, different metrics make different ranking errors. This argues for 4This particular test was chosen since it corrects for multiple post-hoc analyses conducted on the same data-set. the use of multiple metrics in comparing realizer systems. 6 Conclusion Our study suggests that although the task of evaluating the output from realizer systems differs from the task of evaluating machine translations, the automatic metrics used to evaluate MT outputs deliver moderate correlations with combined human fluency and adequacy scores when used on surface realizations. We also found that the MT-evaluation metrics are useful in evaluating different versions of the same realizer system (e.g., the various OpenCCG realization ranking models), and finding cases where a system is performing poorly. As in MT-evaluation tasks, human-targeted metrics have the highest correlations with human judgments overall. These results suggest that the MT-evaluation metrics are useful for developing surface realizers. However, the correlations are lower than those reported for MT data, suggesting that they should be used with caution, especially for cross-system evaluation, where consulting multiple metrics may yield more reliable comparisons. In our study, the targeted version of TERP correlated most strongly with human judgments of fluency. In future work, the performance of the TER family of metrics on this data might be improved by opti- mizing the edit weights used in computing its scores, so as to avoid over-penalizing punctuation movements or under-penalizing agreement errors, both of which were significant sources of ranking errors. Multiple reference sentences may also help mitigate these problems, and the corpus of human-repaired realizations that has resulted from our study is a step in this direction, as it provides multiple references for some cases. We expect the corpus to also prove useful for feature engineering and error analysis in developing better realization models.5 Acknowledgements We thank Aoife Cahill and Tracy King for providing us with the output of the XLE generator. We also thank Chris Callison-Burch and the anonymous reviewers for their helpful comments and suggestions. 5The corpus can be downloaded from http : / /www . l ing .ohio-st ate . edu / ˜mwhite / dat a / emnlp 10 / . 571 This material is based upon work supported by the National Science Foundation under Grant No. 0812297. References Jason Baldridge. 2002. Lexically Specified Derivational Control in Combinatory Categorial Grammar. Ph.D. thesis, University of Edinburgh. S. Banerjee and A. Lavie. 2005. METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72. R. Barzilay and L. Lee. 2003. Learning to paraphrase: An unsupervised approach using multiple-sequence alignment. In proceedings of HLT-NAACL, volume 2003, pages 16–23. Aoife Cahill. 2009. Correlating human and automatic evaluation of a german surface realiser. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, pages 97–100, Suntec, Singapore, August. Association for Computational Linguistics. C. Callison-Burch, M. Osborne, and P. Koehn. 2006. Reevaluating the role of BLEU in machine translation research. In Proceedings of EACL, volume 2006, pages 249–256. Chris Callison-Burch, Cameron Fordyce, Philipp Koehn, Christof Monz, and Josh Schroeder. 2007. (meta-) evaluation ofmachine translation. In StatMT ’07: Proceedings of the Second Workshop on Statistical Machine Translation, pages 136–158, Morristown, NJ, USA. Association for Computational Linguistics. C. Callison-Burch, T. Cohn, and M. Lapata. 2008. Parametric: An automatic evaluation metric for paraphrasing. In Proceedings of the 22nd International Conference on Computational Linguistics-Volume 1, pages 97–104. Association for Computational Linguistics. J. Carletta. 1996. Assessing agreement on classification tasks: the kappa statistic. Computational linguistics, 22(2):249–254. Dick Crouch, Mary Dalrymple, Ron Kaplan, Tracy King, John Maxwell, and Paula Newman. 2008. Xle documentation. Technical report, Palo Alto Research Center. Philip Koenh. 2004. Statistical significance tests for machine translation evaluation. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing. J.R. Landis and G.G. Koch. 1977. The measurement of observer agreement for categorical data. Biometrics, 33(1): 159–174. Lin and Franz Josef Och. 2004. Orange: a method for evaluating automatic evaluation metrics for machine translation. In COLING ’04: Proceedings Chin-Yew of the 20th international conference on Computational 501, Morristown, NJ, USA. Associfor Computational Linguistics. Papineni, S. Roukos, T. Ward, and W. Zhu. 2001. Linguistics, page ation K. Bleu: a method for automatic evaluation of machine translation. E. Technical report, IBM Research. Reiter and A. Belz. 2009. An investigation into the validity of some metrics for automatically evaluating natural language generation systems. Computational Linguistics, 35(4):529–558. Stefan Riezler, Tracy H. King, Ronald M. Kaplan, Richard Crouch, John T. III Maxwell, and Mark Johnson. 2002. Parsing the wall street journal using a lexical-functional grammar and discriminative estimation techniques. In Proceedings of 40th Annual Meeting of the Association for Computational Linguistics, pages 271–278, Philadelphia, Pennsylvania, USA, July. Association for Computational Linguistics. Matthew Snover, Bonnie Dorr, Richard Schwartz, Linnea Micciulla, and John Makhoul. 2006. A study of translation edit rate with targeted human annotation. In In Proceedings of Association for Machine Translation in the Americas, pages 223–23 1. M. Snover, N. Madnani, B.J. Dorr, and R. Schwartz. 2009. Fluency, adequacy, or HTER?: exploring different human judgments with a tunable MT metric. In Proceedings of the Fourth Workshop on Statistical Machine Translation, pages 259–268. Association for Computational Linguistics. Amanda Stent, Matthew Marge, and Mohit Singhai. 2005. Evaluating evaluation methods for generation in the presence of variation. In Proceedings of CICLing. J.P. Turian, L. Shen, and I.D. Melamed. 2003. Evaluation of machine translation and its evaluation. recall (C— R), 100:2. Michael White and Rajakrishnan Rajkumar. 2009. Perceptron reranking for CCG realization. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 410–419, Singapore, August. Association for Computational Linguistics. Michael White. 2006. Efficient Realization of Coordinate Structures in Combinatory Categorial Grammar. Research on Language and Computation, 4(1):39–75. 572 Table 1: Descriptive statistics Table 2: Corpora-wise inter-annotator agreement (absolute and relative κ values shown) SXAROWlpeyLos-aErFndAliCzueqrtd-GAFluq0 N.354217690 B.356219470M .35287410G .35241780 TP.465329170T.A34521670T.465230 H.54T76321H0 .543N89270H.653B7491280H.563M41270H.5643G218 Table 3: Spearman’s correlations among NIST (N), BLEU (B), METEOR (M), GTM (G), TERp (TP), TERpa (TA), TER (T), human variants (HN, HB, HM, HT, HG) and human judgments (-Adq: adequacy and -Flu: Fluency); Scores which fall within the 95 %CI of the best are italicized. SROXAWlLeypoasErldniCze rtG0 N.35246 190 B.5618740 M.542719G0 .5341890T .P632180T.A54268 0T .629310 H.7T6 3985H0 .546N180 H.765B8730H.673M5190 H.56G 4318 Table 4: Spearman’s correlations among NIST (N), BLEU (B), METEOR (M), GTM (G), TERp (TP), TERpa (TA), TER (T), human variants (HN, HB, HM, HT, HG) and human judgments (combined adequacy and fluency scores) 573 SRAXOWylLpeosatrEldniezCm rtG0S A.p61d35q94107 .5304%65874L09.5462%136U0SF .lp256u 1209 .51 6%9213L0 .562%91845U Table 5: Spearman’s correlation analysis (bootstrap sampling) of the BLEU scores of various systems with human adequacy and fluency scores SRXOAWylLpeosarEndiCztGH J -12 0 N.6543210 B.6512830 M.4532 960 G.13457960T.P56374210T.A45268730T.562738140 H.7T6854910H.56N482390H.675B1398240H.567M3 240H.56G41290H.8J71562- Table 6: Spearman’s correlations of NIST (N), BLEU (B), METEOR (M), GTM (G), TERp (TP), TERpa (TA), human variants (HT, HN, HB, HM, HG), and individual human judgments (combined adq. and flu. scores) Factor Count Punctuation17 Adverbial shift Agreement Other shifts Conjunct rearrangement Complementizer ins/del PP shift 16 14 8 8 5 4 Table 7: Factors influencing TERP ranking errors for 50 worst-ranked realization groups Table 8: Metric-wise ranking performance in terms of agreement with a ranking induced by combined adequacy and fluency scores; each metric gets a score out of 5 (i.e. number of system-level comparisons that emerged significant as per the Tukey’s HSD test) Legend: Perceptron Best (PB); Perceptron Worst (PW); XLE Best (XB); XLE Worst (XW); OpenCCG baseline models 1 to 3 (C1 ... C3) 574
6 0.047432486 43 emnlp-2010-Enhancing Domain Portability of Chinese Segmentation Model Using Chi-Square Statistics and Bootstrapping
7 0.04286094 34 emnlp-2010-Crouching Dirichlet, Hidden Markov Model: Unsupervised POS Tagging with Context Local Tag Generation
8 0.042740621 2 emnlp-2010-A Fast Decoder for Joint Word Segmentation and POS-Tagging Using a Single Discriminative Model
9 0.040915802 25 emnlp-2010-Better Punctuation Prediction with Dynamic Conditional Random Fields
10 0.040002339 39 emnlp-2010-EMNLP 044
11 0.039643776 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
12 0.035901647 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
14 0.033048641 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
15 0.032654624 116 emnlp-2010-Using Universal Linguistic Knowledge to Guide Grammar Induction
16 0.032306891 106 emnlp-2010-Top-Down Nearly-Context-Sensitive Parsing
17 0.031910174 19 emnlp-2010-Automatic Analysis of Rhythmic Poetry with Applications to Generation and Translation
18 0.031222077 20 emnlp-2010-Automatic Detection and Classification of Social Events
19 0.029819991 118 emnlp-2010-Utilizing Extra-Sentential Context for Parsing
20 0.029686714 51 emnlp-2010-Function-Based Question Classification for General QA
topicId topicWeight
[(0, 0.121), (1, 0.042), (2, -0.009), (3, -0.013), (4, -0.062), (5, 0.029), (6, 0.014), (7, -0.026), (8, -0.038), (9, -0.039), (10, -0.024), (11, -0.036), (12, 0.054), (13, -0.048), (14, 0.026), (15, 0.055), (16, 0.087), (17, -0.065), (18, -0.06), (19, -0.071), (20, 0.039), (21, -0.094), (22, 0.103), (23, 0.005), (24, 0.001), (25, -0.016), (26, 0.163), (27, -0.04), (28, -0.066), (29, 0.143), (30, -0.076), (31, 0.27), (32, -0.053), (33, -0.075), (34, -0.017), (35, -0.109), (36, -0.129), (37, 0.055), (38, -0.079), (39, 0.107), (40, 0.187), (41, 0.491), (42, 0.038), (43, -0.152), (44, 0.029), (45, -0.034), (46, 0.093), (47, 0.033), (48, 0.021), (49, -0.079)]
simIndex simValue paperId paperTitle
same-paper 1 0.98035258 13 emnlp-2010-A Simple Domain-Independent Probabilistic Approach to Generation
Author: Gabor Angeli ; Percy Liang ; Dan Klein
Abstract: Percy Liang UC Berkeley Berkeley, CA 94720 pliang@cs.berkeley.edu Dan Klein UC Berkeley Berkeley, CA 94720 klein@cs.berkeley.edu We operate in a setting in which we are only given examples consisting of (i) a set of database records We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. We deployed our system in three different domains—Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-ofthe-art domain-specific systems both in terms of BLEU scores and human evaluation.
Author: Aida Khemakhem ; Bilel Gargouri ; Abdelmajid Ben Hamadou
Abstract: Electronic dictionaries covering all natural language levels are very relevant for the human use as well as for the automatic processing use, namely those constructed with respect to international standards. Such dictionaries are characterized by a complex structure and an important access time when using a querying system. However, the need of a user is generally limited to a part of such a dictionary according to his domain and expertise level which corresponds to a specialized dictionary. Given the importance of managing a unified dictionary and considering the personalized needs of users, we propose an approach for generating personalized views starting from a normalized dictionary with respect to Lexical Markup Framework LMF-ISO 24613 norm. This approach provides the re-use of already defined views for a community of users by managing their profiles information and promoting the materialization of the generated views. It is composed of four main steps: (i) the projection of data categories controlled by a set of constraints (related to the user‟s profiles), (ii) the selection of values with consistency checking, (iii) the automatic generation of the query‟s model and finally, (iv) the refinement of the view. The proposed approach was con- solidated by carrying out an experiment on an LMF normalized Arabic dictionary. 1
3 0.32082745 64 emnlp-2010-Incorporating Content Structure into Text Analysis Applications
Author: Christina Sauper ; Aria Haghighi ; Regina Barzilay
Abstract: In this paper, we investigate how modeling content structure can benefit text analysis applications such as extractive summarization and sentiment analysis. This follows the linguistic intuition that rich contextual information should be useful in these tasks. We present a framework which combines a supervised text analysis application with the induction of latent content structure. Both of these elements are learned jointly using the EM algorithm. The induced content structure is learned from a large unannotated corpus and biased by the underlying text analysis task. We demonstrate that exploiting content structure yields significant improvements over approaches that rely only on local context.1
4 0.31981876 99 emnlp-2010-Statistical Machine Translation with a Factorized Grammar
Author: Libin Shen ; Bing Zhang ; Spyros Matsoukas ; Jinxi Xu ; Ralph Weischedel
Abstract: In modern machine translation practice, a statistical phrasal or hierarchical translation system usually relies on a huge set of translation rules extracted from bi-lingual training data. This approach not only results in space and efficiency issues, but also suffers from the sparse data problem. In this paper, we propose to use factorized grammars, an idea widely accepted in the field of linguistic grammar construction, to generalize translation rules, so as to solve these two problems. We designed a method to take advantage of the XTAG English Grammar to facilitate the extraction of factorized rules. We experimented on various setups of low-resource language translation, and showed consistent significant improvement in BLEU over state-ofthe-art string-to-dependency baseline systems with 200K words of bi-lingual training data.
5 0.25263622 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
Author: Xian Qian ; Qi Zhang ; Yaqian Zhou ; Xuanjing Huang ; Lide Wu
Abstract: Many sequence labeling tasks in NLP require solving a cascade of segmentation and tagging subtasks, such as Chinese POS tagging, named entity recognition, and so on. Traditional pipeline approaches usually suffer from error propagation. Joint training/decoding in the cross-product state space could cause too many parameters and high inference complexity. In this paper, we present a novel method which integrates graph structures of two subtasks into one using virtual nodes, and performs joint training and decoding in the factorized state space. Experimental evaluations on CoNLL 2000 shallow parsing data set and Fourth SIGHAN Bakeoff CTB POS tagging data set demonstrate the superiority of our method over cross-product, pipeline and candidate reranking approaches.
6 0.24448755 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
7 0.23135519 4 emnlp-2010-A Game-Theoretic Approach to Generating Spatial Descriptions
9 0.20919921 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
10 0.20551413 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
11 0.19570531 25 emnlp-2010-Better Punctuation Prediction with Dynamic Conditional Random Fields
12 0.18427585 118 emnlp-2010-Utilizing Extra-Sentential Context for Parsing
13 0.17017214 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
14 0.16894476 34 emnlp-2010-Crouching Dirichlet, Hidden Markov Model: Unsupervised POS Tagging with Context Local Tag Generation
15 0.16481644 9 emnlp-2010-A New Approach to Lexical Disambiguation of Arabic Text
16 0.16413662 46 emnlp-2010-Evaluating the Impact of Alternative Dependency Graph Encodings on Solving Event Extraction Tasks
17 0.16377465 39 emnlp-2010-EMNLP 044
18 0.16028966 20 emnlp-2010-Automatic Detection and Classification of Social Events
19 0.15235823 77 emnlp-2010-Measuring Distributional Similarity in Context
20 0.1491124 101 emnlp-2010-Storing the Web in Memory: Space Efficient Language Models with Constant Time Retrieval
topicId topicWeight
[(3, 0.011), (4, 0.014), (10, 0.013), (12, 0.027), (29, 0.113), (30, 0.014), (32, 0.014), (33, 0.347), (51, 0.024), (52, 0.038), (56, 0.079), (62, 0.012), (66, 0.084), (72, 0.054), (76, 0.028), (77, 0.028), (87, 0.014)]
simIndex simValue paperId paperTitle
same-paper 1 0.78464156 13 emnlp-2010-A Simple Domain-Independent Probabilistic Approach to Generation
Author: Gabor Angeli ; Percy Liang ; Dan Klein
Abstract: Percy Liang UC Berkeley Berkeley, CA 94720 pliang@cs.berkeley.edu Dan Klein UC Berkeley Berkeley, CA 94720 klein@cs.berkeley.edu We operate in a setting in which we are only given examples consisting of (i) a set of database records We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. We deployed our system in three different domains—Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-ofthe-art domain-specific systems both in terms of BLEU scores and human evaluation.
2 0.43407649 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
Author: Kristian Woodsend ; Yansong Feng ; Mirella Lapata
Abstract: The task of selecting information and rendering it appropriately appears in multiple contexts in summarization. In this paper we present a model that simultaneously optimizes selection and rendering preferences. The model operates over a phrase-based representation of the source document which we obtain by merging PCFG parse trees and dependency graphs. Selection preferences for individual phrases are learned discriminatively, while a quasi-synchronous grammar (Smith and Eisner, 2006) captures rendering preferences such as paraphrases and compressions. Based on an integer linear programming formulation, the model learns to generate summaries that satisfy both types of preferences, while ensuring that length, topic coverage and grammar constraints are met. Experiments on headline and image caption generation show that our method obtains state-of-the-art performance using essentially the same model for both tasks without any major modifications.
3 0.43229929 116 emnlp-2010-Using Universal Linguistic Knowledge to Guide Grammar Induction
Author: Tahira Naseem ; Harr Chen ; Regina Barzilay ; Mark Johnson
Abstract: We present an approach to grammar induction that utilizes syntactic universals to improve dependency parsing across a range of languages. Our method uses a single set of manually-specified language-independent rules that identify syntactic dependencies between pairs of syntactic categories that commonly occur across languages. During inference of the probabilistic model, we use posterior expectation constraints to require that a minimum proportion of the dependencies we infer be instances of these rules. We also automatically refine the syntactic categories given in our coarsely tagged input. Across six languages our approach outperforms state-of-theart unsupervised methods by a significant margin.1
4 0.43012011 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
Author: Adria de Gispert ; Juan Pino ; William Byrne
Abstract: We report on investigations into hierarchical phrase-based translation grammars based on rules extracted from posterior distributions over alignments of the parallel text. Rather than restrict rule extraction to a single alignment, such as Viterbi, we instead extract rules based on posterior distributions provided by the HMM word-to-word alignmentmodel. We define translation grammars progressively by adding classes of rules to a basic phrase-based system. We assess these grammars in terms of their expressive power, measured by their ability to align the parallel text from which their rules are extracted, and the quality of the translations they yield. In Chinese-to-English translation, we find that rule extraction from posteriors gives translation improvements. We also find that grammars with rules with only one nonterminal, when extracted from posteri- ors, can outperform more complex grammars extracted from Viterbi alignments. Finally, we show that the best way to exploit source-totarget and target-to-source alignment models is to build two separate systems and combine their output translation lattices.
5 0.42931095 82 emnlp-2010-Multi-Document Summarization Using A* Search and Discriminative Learning
Author: Ahmet Aker ; Trevor Cohn ; Robert Gaizauskas
Abstract: In this paper we address two key challenges for extractive multi-document summarization: the search problem of finding the best scoring summary and the training problem of learning the best model parameters. We propose an A* search algorithm to find the best extractive summary up to a given length, which is both optimal and efficient to run. Further, we propose a discriminative training algorithm which directly maximises the quality ofthe best summary, rather than assuming a sentence-level decomposition as in earlier work. Our approach leads to significantly better results than earlier techniques across a number of evaluation metrics.
7 0.42916426 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
8 0.42875007 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
9 0.42862606 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
10 0.42654657 34 emnlp-2010-Crouching Dirichlet, Hidden Markov Model: Unsupervised POS Tagging with Context Local Tag Generation
11 0.42426375 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
12 0.42338276 87 emnlp-2010-Nouns are Vectors, Adjectives are Matrices: Representing Adjective-Noun Constructions in Semantic Space
13 0.42239746 89 emnlp-2010-PEM: A Paraphrase Evaluation Metric Exploiting Parallel Texts
14 0.42148331 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
15 0.42138809 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
16 0.41975799 103 emnlp-2010-Tense Sense Disambiguation: A New Syntactic Polysemy Task
17 0.41717067 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
18 0.41465178 67 emnlp-2010-It Depends on the Translation: Unsupervised Dependency Parsing via Word Alignment
19 0.41425338 60 emnlp-2010-Improved Fully Unsupervised Parsing with Zoomed Learning
20 0.41363233 58 emnlp-2010-Holistic Sentiment Analysis Across Languages: Multilingual Supervised Latent Dirichlet Allocation