| { |
| "paper_id": "2022", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T03:29:11.368571Z" |
| }, |
| "title": "Perceptual Quality Dimensions of Machine-Generated Text with a Focus on Machine Translation", |
| "authors": [ |
| { |
| "first": "Vivien", |
| "middle": [], |
| "last": "Macketanz", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "vivien.macketanz@dfki.de" |
| }, |
| { |
| "first": "Babak", |
| "middle": [], |
| "last": "Naderi", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "Quality and Usability Lab", |
| "institution": "", |
| "location": { |
| "settlement": "TU Berlin" |
| } |
| }, |
| "email": "babak.naderi@tu-berlin.de" |
| }, |
| { |
| "first": "Steven", |
| "middle": [], |
| "last": "Schmidt", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "Quality and Usability Lab", |
| "institution": "", |
| "location": { |
| "settlement": "TU Berlin" |
| } |
| }, |
| "email": "steven.schmidt@tu-berlin.de" |
| }, |
| { |
| "first": "Sebastian", |
| "middle": [], |
| "last": "M\u00f6ller", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "Quality and Usability Lab", |
| "institution": "", |
| "location": { |
| "settlement": "TU Berlin" |
| } |
| }, |
| "email": "sebastian.moeller@tu-berlin.de" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The quality of machine-generated text is a complex construct consisting of various aspects and dimensions. We present a study that aims to uncover relevant perceptual quality dimensions for one type of machine-generated text, that is, Machine Translation. We conducted a crowdsourcing survey in the style of a Semantic Differential to collect attribute ratings for German MT outputs. An Exploratory Factor Analysis revealed the underlying perceptual dimensions. As a result, we extracted four factors that operate as relevant dimensions for the Quality of Experience of MT outputs: precision, complexity, grammaticality, and transparency.", |
| "pdf_parse": { |
| "paper_id": "2022", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The quality of machine-generated text is a complex construct consisting of various aspects and dimensions. We present a study that aims to uncover relevant perceptual quality dimensions for one type of machine-generated text, that is, Machine Translation. We conducted a crowdsourcing survey in the style of a Semantic Differential to collect attribute ratings for German MT outputs. An Exploratory Factor Analysis revealed the underlying perceptual dimensions. As a result, we extracted four factors that operate as relevant dimensions for the Quality of Experience of MT outputs: precision, complexity, grammaticality, and transparency.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "In recent years, automatically generated text has increasingly gained importance, e.g., chatbots, automatic summarizations, or machine translations. Although the quality of such texts has greatly improved over time, it has not yet reached human parity (Toral et al., 2018) . Therefore, the quality of machine-generated text is of ongoing interest to the research community and is further important for gaining acceptance in different applications.", |
| "cite_spans": [ |
| { |
| "start": 252, |
| "end": 272, |
| "text": "(Toral et al., 2018)", |
| "ref_id": "BIBREF23" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The Quality of Experience (QoE) is defined as \"the degree of delight or annoyance of the user of an application or service\" (Le Callet et al., 2012) . This means that the QoE is a subjective perception that needs to be quantified in empirical studies (M\u00f6ller and Raake, 2014) . While there are standardized methods for auditory and visual media, such as ITU P.800, P.910, or BT.500, the QoE of text has been mostly disregarded until now.", |
| "cite_spans": [ |
| { |
| "start": 128, |
| "end": 148, |
| "text": "Callet et al., 2012)", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 251, |
| "end": 275, |
| "text": "(M\u00f6ller and Raake, 2014)", |
| "ref_id": "BIBREF14" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The perceptual quality of machine-generated text is a highly complex construct. Many aspects and dimensions play a crucial role; hence, it is the object of investigation of various research areas. We suggest that a multi-dimensional prediction model covering a wide variety of aspects is the best approach to assess the quality of machine-generated text. To the best of our knowledge, no such model exists. Therefore, we are developing a prediction model for the quality of German machine-generated text, specifically, Machine Translation (MT). We aim to create our model based on a combination of linguistic data and automatically extractable factors that can predict the QoE of MT outputs. Our first milestone is identifying relevant perceptual quality dimensions, the foundation of our model. We achieved this milestone by conducting a crowdsourcing study in the style of a Semantic Differential and subsequently extracting the quality dimensions through an Exploratory Factor Analysis.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "This section provides an overview of the existing metrics for capturing the performance or quality of MT systems. The first category of metrics is automatic methods, which have the advantage of being fast, low-cost, and reproducible. The most commonly used metrics are BLEU (Papineni et al., 2002) , METEOR (Banerjee and Lavie, 2005) , COMET (Rei et al., 2020) , and PRISM (Thompson and Post, 2020) . Metrics like TER (Snover et al., 2009) measure the translation edit rate, and quality estimation methods (Blatz et al., 2004; Specia et al., 2009) can predict the quality without access to the reference translation(s). However, one shared shortcoming of all these automatic metrics is that, as opposed to our approach, they are not based on relevant quality dimensions and thus lack diagnostic power.", |
| "cite_spans": [ |
| { |
| "start": 274, |
| "end": 297, |
| "text": "(Papineni et al., 2002)", |
| "ref_id": "BIBREF18" |
| }, |
| { |
| "start": 307, |
| "end": 333, |
| "text": "(Banerjee and Lavie, 2005)", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 342, |
| "end": 360, |
| "text": "(Rei et al., 2020)", |
| "ref_id": "BIBREF19" |
| }, |
| { |
| "start": 373, |
| "end": 398, |
| "text": "(Thompson and Post, 2020)", |
| "ref_id": "BIBREF22" |
| }, |
| { |
| "start": 418, |
| "end": 439, |
| "text": "(Snover et al., 2009)", |
| "ref_id": "BIBREF20" |
| }, |
| { |
| "start": 506, |
| "end": 526, |
| "text": "(Blatz et al., 2004;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 527, |
| "end": 547, |
| "text": "Specia et al., 2009)", |
| "ref_id": "BIBREF21" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The second category of metrics is subjective methods for directly measuring quality that are more costly yet more reliable. There are largescale human rankings that are often conducted in international conferences in order to compare the performance and/or quality of several MT systems (Callison-Burch et al., 2007; Bojar et al., 2015) . The Multidimensional Quality Metrics (MQM) is a framework for the manual assessment of translation quality (Lommel et al., 2014b) . Additionally, test suites have recently regained more importance. A test suite is a challenge set created to systematically analyze the behavior of MT systems in different aspects, e.g., (Guillou and Hardmeier, 2016) , (Isabelle et al., 2017) , or (Burchardt et al., 2017) .", |
| "cite_spans": [ |
| { |
| "start": 287, |
| "end": 316, |
| "text": "(Callison-Burch et al., 2007;", |
| "ref_id": "BIBREF6" |
| }, |
| { |
| "start": 317, |
| "end": 336, |
| "text": "Bojar et al., 2015)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 446, |
| "end": 468, |
| "text": "(Lommel et al., 2014b)", |
| "ref_id": "BIBREF13" |
| }, |
| { |
| "start": 658, |
| "end": 687, |
| "text": "(Guillou and Hardmeier, 2016)", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 690, |
| "end": 713, |
| "text": "(Isabelle et al., 2017)", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 719, |
| "end": 743, |
| "text": "(Burchardt et al., 2017)", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "While the mentioned techniques focus on capturing the performance or quality of MT systems, they cannot sufficiently capture the QoE by users of MT output as QoE is the only technique that is not measured by pre-defined criteria. Instead, QoE is based on identifying relevant criteria (i.e., quality dimensions) in a real-world scenario.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "We conducted a study to identify relevant dimensions for the quality of machine-generated text, specifically German MT outputs. We did so by utilizing a crowdsourcing survey in which participants had to rate MT outputs. Our corpus contained English to German translations from the submissions to the News translation task of the Fourth Conference on Machine Translation (WMT19) 1 . We chose this data for our corpus as we needed test sentences from several MT systems with varying translation quality. Furthermore, the data is freely available for research purposes 2 . We extracted a set of translations from six submitted systems that appeared at the top, the middle, and the bottom of the ranking of WMT19 systems (Barrault et al., 2019) , resulting in a corpus of 11,922 sentences. A linguistic expert created a sub-corpus for the survey, dedicating around 15 hours to carefully extract translations varying in length, quality, and error types. The sub-corpus consists of 45 sentences. 3 The survey was conducted as a Semantic Differential (SD) (Osgood et al., 1957) . An SD is a rating scale that measures a person's attitude towards an entity, here: our test sentences. The participants were asked to rate their perception of the test items on a scale between two polar adjectives, e.g., \"grammatical -ungrammatical\". All adjective pairs used in the study can be found in Table 2 in the Appendix. The adjective pairs were carefully selected by a linguist who is experienced in MT evaluation and thereafter discussed with another linguist to cover all potentially relevant aspects for the perceptual quality of the test sentences.", |
| "cite_spans": [ |
| { |
| "start": 717, |
| "end": 740, |
| "text": "(Barrault et al., 2019)", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 990, |
| "end": 991, |
| "text": "3", |
| "ref_id": null |
| }, |
| { |
| "start": 1049, |
| "end": 1070, |
| "text": "(Osgood et al., 1957)", |
| "ref_id": "BIBREF17" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 1378, |
| "end": 1385, |
| "text": "Table 2", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Experimental Setup", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We would like to emphasize that while we are using MT as an example text type, the focus of our study lies on the quality of machine-generated text. Therefore, we solely work with the MT outputs and do not take the source sentences and concomitant quality aspects into account (as opposed to approaches that focus on the quality of MT).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experimental Setup", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We first ran a small-scale preliminary study with 14 participants to confirm our antonym pairs. The participants were colleagues and mostly linguistic experts. Our test set comprised 15 sentences from the sub-corpus. The first part of the study consisted of the SD; the participants were instructed to rate the quality of each sentence based on 38 adjective pairs serving as endpoints of a 7-point Likert scale ranging from -3 to +3. As we are solely focusing on the intrinsic quality, they were instructed to rate only the quality of the language but not of the translation itself. The adjective pairs were hand-selected by a linguistic expert, experienced in the evaluation of MT, to cover as many aspects of machine-translated text as possible. In the second part, the participants had to rate each adjective pair on its suitability to evaluate language on a 5-point scale. In addition, they were also encouraged to provide feedback regarding the suitability and to suggest other potential adjective pairs. Based on the rating of the adjective pairs, we removed all adjective pairs with a mean value of less than 3.2 and a standard deviation of more than 1.2. As a result, we reduced the number of adjective pairs to 20.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Antonym pair identification study", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "The main study was conducted as a crowdsourcing survey with Crowdee 4 . 141 crowdworkers participated in the study. The survey followed the IRB guidelines of our institution, and participants were paid according to the minimum wage law. Crowdworkers stayed anonymous, no personal information was collected in the survey 5 . The study was accessible to native speakers only as a good knowledge of German was required. As we wanted the participants to evaluate the language itself (and not the content of the test sentences), they were instructed to base their ratings exclusively on the language of the sentences and ignore the meaning of the sentences as best as they could. They were only informed that the sentences might contain errors, but not that the sentences were outputs of English to German MT. The full instructions can be found in Table 3 in the Appendix.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 843, |
| "end": 850, |
| "text": "Table 3", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Crowdsourcing study", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The adjective pairs were randomized per participant, and so was the order of the polarity. All 45 sentences from the sub-corpus were used. While this is a comparably small number of test items, we argue that we can still draw significant conclusions as the items were hand-picked by an expert to cover as many different linguistic aspects as possible. Based on the feedback we received from the preliminary study, we decided to present only three test sentences to each participant, as the rating is very time-consuming. Each sentence had to be rated based on all 20 antonym pairs. Completing the full survey was expected to take around 10 minutes.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Crowdsourcing study", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "Following (Naderi et al., 2015) , we incorporated a test condition for the majority of the sentences 6 . The test condition is based on calculating an Inconsistency Score (IS) (Naderi, 2018) on repeated adjective pairs. Altogether, we collected up to 30 ratings of all adjective pairs per sentence. The average working time amounted to 392.1 seconds.", |
| "cite_spans": [ |
| { |
| "start": 10, |
| "end": 31, |
| "text": "(Naderi et al., 2015)", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 176, |
| "end": 190, |
| "text": "(Naderi, 2018)", |
| "ref_id": "BIBREF15" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Crowdsourcing study", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "QoE can be formalized as a multidimensional perceptual space where the defining parameters function as dimensions. It is the aim of the multidimensional analysis to identify those dimensions for the QoE of MT output.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multidimensional Analysis", |
| "sec_num": "4" |
| }, |
| { |
| "text": "While crowdsourcing studies have many benefits, one shortcoming is that there might be crowdworkers who do not work thoroughly, eventually leading to noisy data (Naderi et al., 2015) . Thus, we had to cleanse the data to filter out invalid ratings. 7 We did so in three steps: First, we eliminated ratings of participants that completed the survey in 40% or less of the expected 10 minutes. Thus, participants who finished the questionnaire in 240 seconds or less were excluded from the analysis. Second, we excluded all ratings of participants who provided the same value for every adjective pair for every sentence, assuming they were not reading the test material. Lastly, we calculated the IS (Naderi, 2018) . While it is known that the degree of variance in human evaluation of translation is high (Lommel et al., 2014a) , the IS allows filtering out outliers that show a higher degree of variance than expected under normal conditions. The IS calculation is based on the test conditions of the repeated adjective pairs. For details of the calculation, the interested reader is referred to Naderi (2018) .", |
| "cite_spans": [ |
| { |
| "start": 161, |
| "end": 182, |
| "text": "(Naderi et al., 2015)", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 249, |
| "end": 250, |
| "text": "7", |
| "ref_id": null |
| }, |
| { |
| "start": 697, |
| "end": 711, |
| "text": "(Naderi, 2018)", |
| "ref_id": "BIBREF15" |
| }, |
| { |
| "start": 803, |
| "end": 825, |
| "text": "(Lommel et al., 2014a)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 1095, |
| "end": 1108, |
| "text": "Naderi (2018)", |
| "ref_id": "BIBREF15" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Data cleansing", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "The data cleansing removed 6,800 ratings, resulting in 14,200 ratings. The average working time after the data cleansing amounted to 473.31 sec.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Data cleansing", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "We conducted an Exploratory Factor Analysis (EFA) in SPSS (IBM Corp.). Factor analysis is a technique for identifying common factors (i.e., latent variables) that explain the correlation among a set of observed variables. The extraction method used was Maximum Likelihood; The rotation method was PROMAX with Kaiser Normalization, leading to non-orthogonal dimensions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Exploratory Factor Analysis", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "It is important to balance the statistical goodnessof-fit and the interpretability of the resulting dimensions (W\u00e4ltermann et al., 2010) . Our data contained several adjective pairs with low communalities and/or cross-loadings differing by less than 0.2. Our interpretation is that these pairs are not specific enough or are related to other, irrelevant aspects. Thus, we removed those attributes for the sake of interpretability. The dimension reduction revealed four factors for eight polar adjective pairs. Pearson's chi-squared test for the goodness of fit was p = 0.36 (\u03c7 2 = 2.06, df = 2). The Kaiser-Meyer-Olkin value was quite high at 0.901, indicating that the data is adequate for a factor analysis.", |
| "cite_spans": [ |
| { |
| "start": 111, |
| "end": 136, |
| "text": "(W\u00e4ltermann et al., 2010)", |
| "ref_id": "BIBREF24" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Exploratory Factor Analysis", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "The distribution of the adjective pairs on the four factors and the explained percentage of variance can be seen in Table 1 . Note that the adjectives are translated into English for better understanding. The four adjective pairs unambiguous -ambiguous (German: eindeutig -mehrdeutig), precisevague (pr\u00e4zise -ungenau), complete -incomplete (vollst\u00e4ndig -l\u00fcckenhaft), and clear -chaotic (klar -wirr) are loading on factor 1 (F1). F1 explains 53.2% of the variance. Factor 2 (F2) is loaded by the two adjective pairs direct -ponderous (direktumst\u00e4ndlich) and simple -complicated (einfachkompliziert) and explains an additional 8.4% of the variance. Only one adjective pair is loading on Fac- tor 3 (F3): grammatical -ungrammatical (grammatisch -ungrammatisch) and another 10.5% of the variance is explained by F3. The fourth factor (F4) is also loaded by one adjective pair only, namely neat -confusing (\u00fcbersichtlich -verwirrend), and it explains an additional 8.0% of the variance. The adjective pairs loading on F1 are all describing characteristics related to precision; hence, this factor is labeled precision. The adjective pairs loading on F2 are related to complexity; thus, F2 is labeled complexity. F3 is labeled grammaticality, and F4 is labeled transparency. The precision and transparency factors seem to overlap while the remaining factors are more easily separable in their meaning.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 116, |
| "end": 123, |
| "text": "Table 1", |
| "ref_id": "TABREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Exploratory Factor Analysis", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Former commonly used quality aspects for MT were fluency and adequacy (cf., e.g., the MQM metrics mentioned in Section 2). While our study has not tested for extrinsic adequacy, as we only presented the MT outputs and not the source sentences, other authors have already stated that fluency is not the central problem in MT nowadays (Bentivogli et al., 2016) . Neural MT has become more fluent, with MT errors being more subtle and thus harder to spot. Our study confirms this claim as the analysis has brought out four other relevant quality dimensions: precision, complexity, grammaticality, and transparency. Interestingly, our 20 antonym pairs did include the adjective pair fluentnon-fluent, as we covered a wide variety of translation issues. However, we had to eliminate this pair during the EFA due to discriminant validity issues.", |
| "cite_spans": [ |
| { |
| "start": 333, |
| "end": 358, |
| "text": "(Bentivogli et al., 2016)", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Quality dimensions", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "Looking at our four dimensions, the factor precision seems to refer to the clarity and completeness of the text. The factor complexity presumably refers to the textual complexity, and sentences with a high rating for the adjectives complicated and ponderous in our study generally tend to be longer. More interesting findings arise when looking further into our data: Sentences with a high rating for the factor grammaticality tend to miss words, contain spelling or punctuation errors, or hold mistranslations. Interestingly though, these sentences tend to be shorter rather than longer. Our theory is that the longer and therefore more convoluted a sentence is, the more difficult it is to spot grammar errors, and, consequently, other factors like complexity become more relevant. Our last dimension, transparency, seems less tangible than the other dimensions. We theorize that it refers to the lucidity of the text. It seems similar to precision, and there is indeed a higher correlation (0.748).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Quality dimensions", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "As a final remark, we would like to point out that the identification of the dimensions in the multidimensional analysis is strongly dependent on the data (W\u00e4ltermann et al., 2010) , i.e., the choice of test sentences and antonym pairs. While we collected a large number of data points, validating these is the subject of future work. Hence, we cannot guarantee that the identified quality dimensions cover all potential perceptions completely. Furthermore, as the survey was conducted with German native speakers, the majority of the participants can be assumed to be WEIRD participants 8 (Henrich et al., 2010) which leads to a demographic bias. Our findings cannot be assumed to be valid for other languages and/or participant groups.", |
| "cite_spans": [ |
| { |
| "start": 155, |
| "end": 180, |
| "text": "(W\u00e4ltermann et al., 2010)", |
| "ref_id": "BIBREF24" |
| }, |
| { |
| "start": 590, |
| "end": 612, |
| "text": "(Henrich et al., 2010)", |
| "ref_id": "BIBREF8" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Quality dimensions", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "We present a study exploring the relevant quality dimensions for MT outputs. We identified antonym pairs of a Semantic Differential in a preliminary study and used these attributes to rate 45 German test sentences. We then carried out an Exploratory Factor Analysis that resulted in the extraction of four relevant quality dimensions: precision, complexity, grammaticality, and transparency. According to our study, these are the quality dimensions that are relevant for the QoE, i.e., the subjective perception of a user of a text.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Outlook", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Our ultimate goal is to develop a prediction model to assess the quality of machine-generated text. We focus on two text types: Machine Translation and Automatic Text Summarization (ATS). Our next step is to identify the relevant quality dimensions for ATS. To do so, we are currently conducting another crowdsourcing study with an adapted set of adjective pairs. The focus on two different types of machine-generated texts allows us to compare the (potential) differences in the perceptive quality dimensions and enables us to draw generalizations for other text types.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Outlook", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Simultaneously, we are working on the quantification of the quality dimensions for MT. As the factor analysis conducted in the study at hand is highly complex, we are developing a simplified survey in which we present only one representative antonym pair per dimension. If the result of the follow-up study verifies our current study, we can assume our dimensions to be accurate.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Outlook", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Further steps will involve correlating automatically extractable text parameters and quality dimensions, and building and testing various prediction models. These efforts should ultimately result in a quality prediction model for MT, ATS, and potentially other types of machine-generated text.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Outlook", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Other potential future work includes analyzing the possible overlap between the four dimensions at hand and other existing quality metrics, e.g., MQM. Furthermore, it would be of interest to expand the analysis to other languages, as it might also counteract the WEIRD bias.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Outlook", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Willkommen zur Umfrage In dieser Umfrage sollst du die Sprache von verschiedenen S\u00e4tzen anhand einer Adjektivliste bewerten. Hierzu werden dir insgesamt 3 S\u00e4tze auf je 4 Seiten gezeigt. Die S\u00e4tze k\u00f6nnen fehlerhaft sein, m\u00fcssen aber nicht. Bitte bewerte jeden dieser 3 S\u00e4tze in Hinblick auf die verwendete Sprache (inklusive Satzzeichen) mit Hilfe der Adjektivliste. Die Adjektivliste enth\u00e4lt 22 geges\u00e4tzliche Adjektivpaare, die an den beiden Enden einer Skala von -3 bis +3 stehen. Bitte schiebe f\u00fcr jedes Adjektivpaar den Slider auf der Skala dorthin, wo der Wert deiner Meinung nach die Sprache des jeweiligen Satzes am besten beschreibt. Versuche, den Inhalt der S\u00e4tze nicht in deine Bewertung miteinflie\u00dfen zu lassen. Alle deine Antworten aus dem folgenden Fragebogen werden anonym behandelt und dienen ausschlie\u00dflich dem Zweck dieser wissenschaftlichen Arbeit. Achtung: Das Ergebnis dieser Umfrage ist sehr wichtig f\u00fcr uns und andere Wissenschaftler, die in diesem Bereich arbeiten. Wir verf\u00fcgen \u00fcber Methoden um die Einheitlichkeit deiner Antworten zu \u00fcberpr\u00fcfen. Wir werden diese Methoden nutzen, um die Qualit\u00e4t der abgeschickten Aufgaben zu bewerten. Crowdworker, die qualitativ hochwertige Antworten geben, werden zu weiteren Untersuchungen eingeladen, zu denen sie exklusiven Zugang erhalten. Auf der n\u00e4chsten Seite wirst du zun\u00e4chst ein Beispiel sehen, bevor es losgeht.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "German original", |
| "sec_num": null |
| }, |
| { |
| "text": "Welcome to the survey In this survey, you are supposed to evaluate the language of different sentences with the help of an adjective list. You will be shown 3 sentences altogether, distributed over 4 pages each. Die sentences might, but don't have to, contain errors. Please evaluate each of the 3 sentences with regard to the language used (including punctuation) with the help of the adjective list. The adjective list contains 22 polar adjective pairs which are located on both ends of a scale from -3 to +3. Please move the slider for each adjective pair to the point on the scale where the value describes the language of the respective sentence best in your opinion. Try to not let the content of the sentences influence your evaluation. All your answers in the following survey will be handled anonymously and exclusively serve the aim of this scientific work. Note: The result of this survey is very important for us and other scientists working in this area. We are equipped with methods to check your answers for consistency. We will use these methods to evaluate the quality of the completed task. Crowdworkers that provide high-quality answers will be invited to further surveys to which they will receive exclusive access. On the next page, you will first see an example before the survey starts. ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "English translation", |
| "sec_num": null |
| }, |
| { |
| "text": "http://www.statmt.org/wmt19/index. html 2 cf. Licensing of Data https://www.statmt.org/ wmt19/translation-task.html 3 https://github.com/DFKI-NLP/TextQ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://www.crowdee.com/ 5 Crowdee's privacy Statement can be found here: https: //www.crowdee.com/privacy-statement", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "30 of the 45 test sentences were rated with the test condition, as we ran the survey in two batches and included the test condition only in the second batch.7 Crowdworkers were paid regardless of their ratings.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "WEIRD stands for western, educated, industrialized, rich, and democratic participants", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "The present study was funded by the Deutsche Forschungsgemeinschaft (DFG) through the project \"Analyse und automatische Absch\u00e4tzung der Qualit\u00e4t maschinell generierter Texte\", project number 436813723. We thank all colleagues who participated in our preliminary study.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgements", |
| "sec_num": null |
| }, |
| { |
| "text": "English translationGroup 1: final list of adjective pairs that are loading on the underlying factors direkt -umst\u00e4ndlich direct -ponderous eindeutig -mehrdeutig unambiguous -ambiguous einfach -kompliziert simple -complicated grammatisch -ungrammatisch grammatical -ungrammatical klar -wirr clear -chaotic pr\u00e4zise -ungenau precise -vague \u00fcbersichtlich -verwirrend neat -confusing vollst\u00e4ndig -l\u00fcckenhaft complete -incompleteGroup 2: list of adjective pairs that were removed during the factor analysis for the sake of interpretability fl\u00fcssig -holprig fluent -non-fluent formell -informell formal -informal geordnet -durcheinander orderly -messy geschrieben -gesprochen written -spoken h\u00f6flich -unh\u00f6flich polite -impolite kongruent -inkongruent congruent -incongruent konsistent -inkonsistent consistent -inconsistent logisch -unlogisch logical -illogical menschlich -technisch human -technical muttersprachlich -fremdprachlich native -foreign-language pers\u00f6nlich -unpers\u00f6nlich personal -impersonal professionell -laienhaft professional -unprofessionalGroup 3: list of adjective pairs that were removed after the preliminary study aktiv -passiv active -passive angemessen -unangemessen appropriate -inappropriate angenehm -unangenehm pleasant -unpleasant bedeutungsvoll -bedeutungslos meaningful -meaningless bekannt -unbekannt known -unknown f\u00f6rmlich -l\u00e4ssig formal -casual gebildet -ungebildet educated -uneducated gut -schlecht good -bad hochwertig -minderwertig valuable -poor informativ -nichtssagend informative -bland kreativ -simpel creative -simple lustig -ernst funny -serious optimal -suboptimal optimal -suboptimal praktisch -unpraktisch practical -impractical stilvoll -stillos classy -unclassy vertraut -fremd familiar -foreign vorhersehbar -unberechenbar predictable -unpredictable warm -kalt warm -cold weich -hart soft -hard zweckorientiert -zweckfrei purposeful -purposeless Table 2 : Complete list of polar adjective pairs used in the study in the German original and translated into English for better understanding.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 1892, |
| "end": 1899, |
| "text": "Table 2", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "German original", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Meteor: An automatic metric for MT evaluation with improved correlation with human judgments", |
| "authors": [ |
| { |
| "first": "Satanjeev", |
| "middle": [], |
| "last": "Banerjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Alon", |
| "middle": [], |
| "last": "Lavie", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization", |
| "volume": "", |
| "issue": "", |
| "pages": "65--72", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Satanjeev Banerjee and Alon 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 summariza- tion, pages 65-72.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Santanu Pal, Matt Post, and Marcos Zampieri", |
| "authors": [ |
| { |
| "first": "Lo\u00efc", |
| "middle": [], |
| "last": "Barrault", |
| "suffix": "" |
| }, |
| { |
| "first": "Ond\u0159ej", |
| "middle": [], |
| "last": "Bojar", |
| "suffix": "" |
| }, |
| { |
| "first": "Marta", |
| "middle": [ |
| "R" |
| ], |
| "last": "Costa-Juss\u00e0", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Federmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Mark", |
| "middle": [], |
| "last": "Fishel", |
| "suffix": "" |
| }, |
| { |
| "first": "Yvette", |
| "middle": [], |
| "last": "Graham", |
| "suffix": "" |
| }, |
| { |
| "first": "Barry", |
| "middle": [], |
| "last": "Haddow", |
| "suffix": "" |
| }, |
| { |
| "first": "Matthias", |
| "middle": [], |
| "last": "Huck", |
| "suffix": "" |
| }, |
| { |
| "first": "Philipp", |
| "middle": [], |
| "last": "Koehn", |
| "suffix": "" |
| }, |
| { |
| "first": "Shervin", |
| "middle": [], |
| "last": "Malmasi", |
| "suffix": "" |
| }, |
| { |
| "first": "Christof", |
| "middle": [], |
| "last": "Monz", |
| "suffix": "" |
| }, |
| { |
| "first": "Mathias", |
| "middle": [], |
| "last": "M\u00fcller", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Findings of the 2019 Conference on Machine Translation (WMT19). In Proceedings of the Fourth Conference on Machine Translation", |
| "volume": "2", |
| "issue": "", |
| "pages": "1--61", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lo\u00efc Barrault, Ond\u0159ej Bojar, Marta R. Costa-juss\u00e0, Christian Federmann, Mark Fishel, Yvette Gra- ham, Barry Haddow, Matthias Huck, Philipp Koehn, Shervin Malmasi, Christof Monz, Mathias M\u00fcller, Santanu Pal, Matt Post, and Marcos Zampieri. 2019. Findings of the 2019 Conference on Machine Trans- lation (WMT19). In Proceedings of the Fourth Con- ference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 1-61, Florence, Italy. As- sociation for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Neural versus phrasebased machine translation quality: a case study", |
| "authors": [ |
| { |
| "first": "Luisa", |
| "middle": [], |
| "last": "Bentivogli", |
| "suffix": "" |
| }, |
| { |
| "first": "Arianna", |
| "middle": [], |
| "last": "Bisazza", |
| "suffix": "" |
| }, |
| { |
| "first": "Mauro", |
| "middle": [], |
| "last": "Cettolo", |
| "suffix": "" |
| }, |
| { |
| "first": "Marcello", |
| "middle": [], |
| "last": "Federico", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": { |
| "arXiv": [ |
| "arXiv:1608.04631" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Luisa Bentivogli, Arianna Bisazza, Mauro Cettolo, and Marcello Federico. 2016. Neural versus phrase- based machine translation quality: a case study. arXiv preprint arXiv:1608.04631.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Confidence estimation for machine translation", |
| "authors": [ |
| { |
| "first": "John", |
| "middle": [], |
| "last": "Blatz", |
| "suffix": "" |
| }, |
| { |
| "first": "Erin", |
| "middle": [], |
| "last": "Fitzgerald", |
| "suffix": "" |
| }, |
| { |
| "first": "George", |
| "middle": [], |
| "last": "Foster", |
| "suffix": "" |
| }, |
| { |
| "first": "Simona", |
| "middle": [], |
| "last": "Gandrabur", |
| "suffix": "" |
| }, |
| { |
| "first": "Cyril", |
| "middle": [], |
| "last": "Goutte", |
| "suffix": "" |
| }, |
| { |
| "first": "Alex", |
| "middle": [], |
| "last": "Kulesza", |
| "suffix": "" |
| }, |
| { |
| "first": "Alberto", |
| "middle": [], |
| "last": "Sanchis", |
| "suffix": "" |
| }, |
| { |
| "first": "Nicola", |
| "middle": [], |
| "last": "Ueffing", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "Proceedings of the 20th International Conference on Computational Linguistics, COLING '04", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": { |
| "DOI": [ |
| "10.3115/1220355.1220401" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "John Blatz, Erin Fitzgerald, George Foster, Simona Gan- drabur, Cyril Goutte, Alex Kulesza, Alberto Sanchis, and Nicola Ueffing. 2004. Confidence estimation for machine translation. In Proceedings of the 20th Inter- national Conference on Computational Linguistics, COLING '04, page 315-es, USA. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Findings of the 2015 Workshop on Statistical Machine Translation", |
| "authors": [ |
| { |
| "first": "Ondrej", |
| "middle": [], |
| "last": "Bojar", |
| "suffix": "" |
| }, |
| { |
| "first": "Rajen", |
| "middle": [], |
| "last": "Chatterjee", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Federmann", |
| "suffix": "" |
| }, |
| { |
| "first": "Barry", |
| "middle": [], |
| "last": "Haddow", |
| "suffix": "" |
| }, |
| { |
| "first": "Matthias", |
| "middle": [], |
| "last": "Huck", |
| "suffix": "" |
| }, |
| { |
| "first": "Chris", |
| "middle": [], |
| "last": "Hokamp", |
| "suffix": "" |
| }, |
| { |
| "first": "Philipp", |
| "middle": [], |
| "last": "Koehn", |
| "suffix": "" |
| }, |
| { |
| "first": "Varvara", |
| "middle": [], |
| "last": "Logacheva", |
| "suffix": "" |
| }, |
| { |
| "first": "Christof", |
| "middle": [], |
| "last": "Monz", |
| "suffix": "" |
| }, |
| { |
| "first": "Matteo", |
| "middle": [], |
| "last": "Negri", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ondrej Bojar, Rajen Chatterjee, Christian Federmann, Barry Haddow, Matthias Huck, Chris Hokamp, Philipp Koehn, Varvara Logacheva, Christof Monz, Matteo Negri, et al. 2015. Findings of the 2015 Workshop on Statistical Machine Translation.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "A linguistic evaluation of rule-based, phrase-based, and neural MT engines", |
| "authors": [ |
| { |
| "first": "Aljoscha", |
| "middle": [], |
| "last": "Burchardt", |
| "suffix": "" |
| }, |
| { |
| "first": "Vivien", |
| "middle": [], |
| "last": "Macketanz", |
| "suffix": "" |
| }, |
| { |
| "first": "Jon", |
| "middle": [], |
| "last": "Dehdari", |
| "suffix": "" |
| }, |
| { |
| "first": "Georg", |
| "middle": [], |
| "last": "Heigold", |
| "suffix": "" |
| }, |
| { |
| "first": "Jan-Thorsten", |
| "middle": [], |
| "last": "Peter", |
| "suffix": "" |
| }, |
| { |
| "first": "Philip", |
| "middle": [], |
| "last": "Williams", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "The Prague Bulletin of Mathematical Linguistics", |
| "volume": "108", |
| "issue": "1", |
| "pages": "159--170", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Aljoscha Burchardt, Vivien Macketanz, Jon Dehdari, Georg Heigold, Jan-Thorsten Peter, and Philip Williams. 2017. A linguistic evaluation of rule-based, phrase-based, and neural MT engines. The Prague Bulletin of Mathematical Linguistics, 108(1):159- 170.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "meta-) evaluation of machine translation", |
| "authors": [ |
| { |
| "first": "Chris", |
| "middle": [], |
| "last": "Callison", |
| "suffix": "" |
| }, |
| { |
| "first": "-", |
| "middle": [], |
| "last": "Burch", |
| "suffix": "" |
| }, |
| { |
| "first": "Cameron", |
| "middle": [], |
| "last": "Shaw Fordyce", |
| "suffix": "" |
| }, |
| { |
| "first": "Philipp", |
| "middle": [], |
| "last": "Koehn", |
| "suffix": "" |
| }, |
| { |
| "first": "Christof", |
| "middle": [], |
| "last": "Monz", |
| "suffix": "" |
| }, |
| { |
| "first": "Josh", |
| "middle": [], |
| "last": "Schroeder", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Proceedings of the Second Workshop on Statistical Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "136--158", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Chris Callison-Burch, Cameron Shaw Fordyce, Philipp Koehn, Christof Monz, and Josh Schroeder. 2007. (meta-) evaluation of machine translation. In Pro- ceedings of the Second Workshop on Statistical Ma- chine Translation, pages 136-158.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "PROTEST: A Test Suite for Evaluating Pronouns in Machine Translation", |
| "authors": [ |
| { |
| "first": "Liane", |
| "middle": [], |
| "last": "Guillou", |
| "suffix": "" |
| }, |
| { |
| "first": "Christian", |
| "middle": [], |
| "last": "Hardmeier", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)", |
| "volume": "", |
| "issue": "", |
| "pages": "636--643", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Liane Guillou and Christian Hardmeier. 2016. PROTEST: A Test Suite for Evaluating Pronouns in Machine Translation. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 636-643, Por- toro\u017e, Slovenia. European Language Resources Association (ELRA).", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "The weirdest people in the world?", |
| "authors": [ |
| { |
| "first": "Joseph", |
| "middle": [], |
| "last": "Henrich", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Steven", |
| "suffix": "" |
| }, |
| { |
| "first": "Ara", |
| "middle": [], |
| "last": "Heine", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Norenzayan", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Behavioral and brain sciences", |
| "volume": "33", |
| "issue": "2-3", |
| "pages": "61--83", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Joseph Henrich, Steven J Heine, and Ara Norenzayan. 2010. The weirdest people in the world? Behavioral and brain sciences, 33(2-3):61-83.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "IBM SPSS Statistics for Macintosh", |
| "authors": [ |
| { |
| "first": "", |
| "middle": [], |
| "last": "Ibm Corp", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "IBM Corp. IBM SPSS Statistics for Macintosh. Version 28.0.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "A Challenge Set Approach to Evaluating Machine Translation", |
| "authors": [ |
| { |
| "first": "Pierre", |
| "middle": [], |
| "last": "Isabelle", |
| "suffix": "" |
| }, |
| { |
| "first": "Colin", |
| "middle": [], |
| "last": "Cherry", |
| "suffix": "" |
| }, |
| { |
| "first": "George", |
| "middle": [], |
| "last": "Foster", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Pierre Isabelle, Colin Cherry, and George Foster. 2017. A Challenge Set Approach to Evaluating Machine Translation.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Qualinet white paper on definitions of quality of experience. European network on quality of experience in multimedia systems and services (COST Action IC 1003", |
| "authors": [ |
| { |
| "first": "Patrick", |
| "middle": [ |
| "Le" |
| ], |
| "last": "Callet", |
| "suffix": "" |
| }, |
| { |
| "first": "Sebastian", |
| "middle": [], |
| "last": "M\u00f6ller", |
| "suffix": "" |
| }, |
| { |
| "first": "Andrew", |
| "middle": [], |
| "last": "Perkis", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Patrick Le Callet, Sebastian M\u00f6ller, Andrew Perkis, et al. 2012. Qualinet white paper on definitions of quality of experience. European network on quality of experience in multimedia systems and services (COST Action IC 1003), 3(2012).", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Assessing inter-annotator agreement for translation error annotation", |
| "authors": [ |
| { |
| "first": "Arle", |
| "middle": [], |
| "last": "Lommel", |
| "suffix": "" |
| }, |
| { |
| "first": "Maja", |
| "middle": [], |
| "last": "Popovic", |
| "suffix": "" |
| }, |
| { |
| "first": "Aljoscha", |
| "middle": [], |
| "last": "Burchardt", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "MTE: Workshop on Automatic and Manual Metrics for Operational Translation Evaluation", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Arle Lommel, Maja Popovic, and Aljoscha Burchardt. 2014a. Assessing inter-annotator agreement for trans- lation error annotation. In MTE: Workshop on Auto- matic and Manual Metrics for Operational Transla- tion Evaluation.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Multidimensional quality metrics (MQM): A framework for declaring and describing translation quality metrics", |
| "authors": [ |
| { |
| "first": "Arle", |
| "middle": [], |
| "last": "Lommel", |
| "suffix": "" |
| }, |
| { |
| "first": "Hans", |
| "middle": [], |
| "last": "Uszkoreit", |
| "suffix": "" |
| }, |
| { |
| "first": "Aljoscha", |
| "middle": [], |
| "last": "Burchardt", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Tradum\u00e0tica", |
| "volume": "", |
| "issue": "12", |
| "pages": "455--463", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Arle Lommel, Hans Uszkoreit, and Aljoscha Burchardt. 2014b. Multidimensional quality metrics (MQM): A framework for declaring and describing translation quality metrics. Tradum\u00e0tica, (12):0455-463.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Quality of experience: advanced concepts, applications and methods", |
| "authors": [ |
| { |
| "first": "Sebastian", |
| "middle": [], |
| "last": "M\u00f6ller", |
| "suffix": "" |
| }, |
| { |
| "first": "Alexander", |
| "middle": [], |
| "last": "Raake", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Sebastian M\u00f6ller and Alexander Raake. 2014. Quality of experience: advanced concepts, applications and methods. Springer.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Motivation of workers on microtask crowdsourcing platforms", |
| "authors": [ |
| { |
| "first": "Babak", |
| "middle": [], |
| "last": "Naderi", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Babak Naderi. 2018. Motivation of workers on micro- task crowdsourcing platforms. Springer.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Effect of being observed on the reliability of responses in crowdsourcing micro-task platforms", |
| "authors": [ |
| { |
| "first": "Babak", |
| "middle": [], |
| "last": "Naderi", |
| "suffix": "" |
| }, |
| { |
| "first": "Ina", |
| "middle": [], |
| "last": "Wechsung", |
| "suffix": "" |
| }, |
| { |
| "first": "Sebastian", |
| "middle": [], |
| "last": "M\u00f6ller", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "Seventh International Workshop on Quality of Multimedia Experience (QoMEX)", |
| "volume": "", |
| "issue": "", |
| "pages": "1--2", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Babak Naderi, Ina Wechsung, and Sebastian M\u00f6ller. 2015. Effect of being observed on the reliability of responses in crowdsourcing micro-task platforms. In 2015 Seventh International Workshop on Quality of Multimedia Experience (QoMEX), pages 1-2. IEEE.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "The measurement of meaning", |
| "authors": [ |
| { |
| "first": "Charles", |
| "middle": [ |
| "Egerton" |
| ], |
| "last": "Osgood", |
| "suffix": "" |
| }, |
| { |
| "first": "George", |
| "middle": [ |
| "J" |
| ], |
| "last": "Suci", |
| "suffix": "" |
| }, |
| { |
| "first": "Percy", |
| "middle": [ |
| "H" |
| ], |
| "last": "Tannenbaum", |
| "suffix": "" |
| } |
| ], |
| "year": 1957, |
| "venue": "", |
| "volume": "47", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Charles Egerton Osgood, George J Suci, and Percy H Tannenbaum. 1957. The measurement of meaning. 47. University of Illinois press.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "BLEU: A Method for Automatic Evaluation of Machine Translation", |
| "authors": [ |
| { |
| "first": "Kishore", |
| "middle": [], |
| "last": "Papineni", |
| "suffix": "" |
| }, |
| { |
| "first": "Salim", |
| "middle": [], |
| "last": "Roukos", |
| "suffix": "" |
| }, |
| { |
| "first": "Todd", |
| "middle": [], |
| "last": "Ward", |
| "suffix": "" |
| }, |
| { |
| "first": "Wei-Jing", |
| "middle": [], |
| "last": "Zhu", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL '02", |
| "volume": "", |
| "issue": "", |
| "pages": "311--318", |
| "other_ids": { |
| "DOI": [ |
| "10.3115/1073083.1073135" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. BLEU: A Method for Automatic Evaluation of Machine Translation. In Proceedings of the 40th Annual Meeting on Association for Com- putational Linguistics, ACL '02, page 311-318, USA. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Unbabel's participation in the WMT20 metrics shared task", |
| "authors": [ |
| { |
| "first": "Ricardo", |
| "middle": [], |
| "last": "Rei", |
| "suffix": "" |
| }, |
| { |
| "first": "Craig", |
| "middle": [], |
| "last": "Stewart", |
| "suffix": "" |
| }, |
| { |
| "first": "Ana", |
| "middle": [ |
| "C" |
| ], |
| "last": "Farinha", |
| "suffix": "" |
| }, |
| { |
| "first": "Alon", |
| "middle": [], |
| "last": "Lavie", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "Proceedings of the Fifth Conference on Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "911--920", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. 2020. Unbabel's participation in the WMT20 metrics shared task. In Proceedings of the Fifth Con- ference on Machine Translation, pages 911-920, On- line. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Fluency, adequacy, or HTER? Exploring different human judgments with a tunable MT metric", |
| "authors": [ |
| { |
| "first": "Matthew", |
| "middle": [], |
| "last": "Snover", |
| "suffix": "" |
| }, |
| { |
| "first": "Nitin", |
| "middle": [], |
| "last": "Madnani", |
| "suffix": "" |
| }, |
| { |
| "first": "Bonnie", |
| "middle": [], |
| "last": "Dorr", |
| "suffix": "" |
| }, |
| { |
| "first": "Richard", |
| "middle": [], |
| "last": "Schwartz", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Proceedings of the Fourth Workshop on Statistical Machine Translation", |
| "volume": "", |
| "issue": "", |
| "pages": "259--268", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Matthew Snover, Nitin Madnani, Bonnie Dorr, and Richard 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, Athens, Greece. Association for Computa- tional Linguistics.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Improving the confidence of machine translation quality estimates", |
| "authors": [ |
| { |
| "first": "Lucia", |
| "middle": [], |
| "last": "Specia", |
| "suffix": "" |
| }, |
| { |
| "first": "Craig", |
| "middle": [], |
| "last": "Saunders", |
| "suffix": "" |
| }, |
| { |
| "first": "Marco", |
| "middle": [], |
| "last": "Turchi", |
| "suffix": "" |
| }, |
| { |
| "first": "Zhuoran", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "John", |
| "middle": [], |
| "last": "Shawe-Taylor", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lucia Specia, Craig Saunders, Marco Turchi, Zhuoran Wang, and John Shawe-Taylor. 2009. Improving the confidence of machine translation quality estimates.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "Automatic machine translation evaluation in many languages via zero-shot paraphrasing", |
| "authors": [ |
| { |
| "first": "Brian", |
| "middle": [], |
| "last": "Thompson", |
| "suffix": "" |
| }, |
| { |
| "first": "Matt", |
| "middle": [], |
| "last": "Post", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)", |
| "volume": "", |
| "issue": "", |
| "pages": "90--121", |
| "other_ids": { |
| "DOI": [ |
| "10.18653/v1/2020.emnlp-main.8" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Brian Thompson and Matt Post. 2020. Automatic ma- chine translation evaluation in many languages via zero-shot paraphrasing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Lan- guage Processing (EMNLP), pages 90-121, Online. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF23": { |
| "ref_id": "b23", |
| "title": "Attaining the Unattainable? Reassessing Claims of Human Parity in Neural Machine Translation", |
| "authors": [ |
| { |
| "first": "Antonio", |
| "middle": [], |
| "last": "Toral", |
| "suffix": "" |
| }, |
| { |
| "first": "Sheila", |
| "middle": [], |
| "last": "Castilho", |
| "suffix": "" |
| }, |
| { |
| "first": "Ke", |
| "middle": [], |
| "last": "Hu", |
| "suffix": "" |
| }, |
| { |
| "first": "Andy", |
| "middle": [], |
| "last": "Way", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Antonio Toral, Sheila Castilho, Ke Hu, and Andy Way. 2018. Attaining the Unattainable? Reassessing Claims of Human Parity in Neural Machine Transla- tion.", |
| "links": null |
| }, |
| "BIBREF24": { |
| "ref_id": "b24", |
| "title": "Quality dimensions of narrowband and wideband speech transmission", |
| "authors": [ |
| { |
| "first": "Marcel", |
| "middle": [], |
| "last": "W\u00e4ltermann", |
| "suffix": "" |
| }, |
| { |
| "first": "Alexander", |
| "middle": [], |
| "last": "Raake", |
| "suffix": "" |
| }, |
| { |
| "first": "Sebastian", |
| "middle": [], |
| "last": "M\u00f6ller", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Acta Acustica united with Acustica", |
| "volume": "96", |
| "issue": "6", |
| "pages": "1090--1103", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marcel W\u00e4ltermann, Alexander Raake, and Sebastian M\u00f6ller. 2010. Quality dimensions of narrowband and wideband speech transmission. Acta Acustica united with Acustica, 96(6):1090-1103.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF1": { |
| "content": "<table/>", |
| "type_str": "table", |
| "text": "Loadings of the adjective pairs (English translations) on the factors and % of explained variance.", |
| "num": null, |
| "html": null |
| }, |
| "TABREF2": { |
| "content": "<table/>", |
| "type_str": "table", |
| "text": "Instructions for the crowdsourcing survey in the German original and translated into English for better understanding.", |
| "num": null, |
| "html": null |
| } |
| } |
| } |
| } |
