2018 |
Behnke, Maximiliana; Miceli Barone, {Antonio Valerio} ; Sennrich, Rico; Sosoni, Vilelmini; Naskos, Thanasis; Takoulidou, Eirini; Stasimioti, Maria; van Zaanan, Menno ; Castilho, Sheila; Gaspari, Federico; Georgakopoulou, Panayota; Kordoni, Valia; Egg, Markus; Katia Lida, Improving Machine Translation of Educational Content via Crowdsourcing Inproceedings 11th Edition of the Language Resources and Evaluation Conference, 2018. @inproceedings{5c6a88b29e3646929888930434d8f7b5b, title = {Improving Machine Translation of Educational Content via Crowdsourcing}, author = {Maximiliana Behnke and {Antonio Valerio} {Miceli Barone} and Rico Sennrich and Vilelmini Sosoni and Thanasis Naskos and Eirini Takoulidou and Maria Stasimioti and Menno {van Zaanan} and Sheila Castilho and Federico Gaspari and Panayota Georgakopoulou and Valia Kordoni and Markus Egg and {Katia Lida} Kermanidis}, year = {2018}, date = {2018-01-01}, booktitle = {11th Edition of the Language Resources and Evaluation Conference}, abstract = {The limited availability of in-domain training data is a major issue in the training of application-specific neural machine translationmodels. Professional outsourcing of bilingual data collections is costly and often not feasible. In this paper we analyze the influence of using crowdsourcing as a scalable way to obtain translations of target in-domain data having in mind that the translations can be of a lower quality. We apply crowdsourcing with carefully designed quality controls to create parallel corpora for the educational domain by collecting translations of texts from MOOCs from English to eleven languages, which we then use to fine-tune neural machine translation models previously trained on general-domain data. The results from our research indicate that crowdsourced data collected with proper quality controls consistently yields performance gains over general-domain baseline systems, and systems fine-tuned with pre-existing in-domain corpora.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } The limited availability of in-domain training data is a major issue in the training of application-specific neural machine translationmodels. Professional outsourcing of bilingual data collections is costly and often not feasible. In this paper we analyze the influence of using crowdsourcing as a scalable way to obtain translations of target in-domain data having in mind that the translations can be of a lower quality. We apply crowdsourcing with carefully designed quality controls to create parallel corpora for the educational domain by collecting translations of texts from MOOCs from English to eleven languages, which we then use to fine-tune neural machine translation models previously trained on general-domain data. The results from our research indicate that crowdsourced data collected with proper quality controls consistently yields performance gains over general-domain baseline systems, and systems fine-tuned with pre-existing in-domain corpora. |
Liu, Yang; Lapata, Maria Learning Structured Text Representations Journal Article Transactions of the Association for Computational Linguistics, 6 , pp. 63–76, 2018, ISSN: 2307-387X. @article{8cfe7cfc1bee4bb289bd87c6d7e94b3e, title = {Learning Structured Text Representations}, author = {Yang Liu and Maria Lapata}, issn = {2307-387X}, year = {2018}, date = {2018-01-01}, journal = {Transactions of the Association for Computational Linguistics}, volume = {6}, pages = {63--76}, abstract = {In this paper, we focus on learning structureaware document representations from data without recourse to a discourse parser or additional annotations. Drawing inspiration from recent efforts to empower neural networks with a structural bias (Cheng et al., 2016; Kim et al., 2017), we propose a model that can encode a document while automatically inducing rich structural dependencies. Specifically, we embed a differentiable non-projective parsing algorithm into a neural model and use attention mechanisms to incorporate the structural biases. Experimental evaluations across different tasks and datasets show that the proposed model achieves state-of-the-art results on document modeling tasks while inducing intermediate structures which are both interpretable and meaningful.}, keywords = {}, pubstate = {published}, tppubtype = {article} } In this paper, we focus on learning structureaware document representations from data without recourse to a discourse parser or additional annotations. Drawing inspiration from recent efforts to empower neural networks with a structural bias (Cheng et al., 2016; Kim et al., 2017), we propose a model that can encode a document while automatically inducing rich structural dependencies. Specifically, we embed a differentiable non-projective parsing algorithm into a neural model and use attention mechanisms to incorporate the structural biases. Experimental evaluations across different tasks and datasets show that the proposed model achieves state-of-the-art results on document modeling tasks while inducing intermediate structures which are both interpretable and meaningful. |
Perez-Beltrachini, Laura; Lapata, Maria Bootstrapping Generators from Noisy Data Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. Abstract | Links | BibTeX | Tags: @inproceedings{16966526dbb3461d9547b3cdc610277a, title = {Bootstrapping Generators from Noisy Data}, author = {Laura Perez-Beltrachini and Maria Lapata}, doi = {10.18653/v1/N18-1137}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {A core step in statistical data-to-text generation concerns learning correspondences between structured data representations (e.g., facts in a database) and associated texts. In this paper we aim to bootstrap generators from large scale datasets where the data (e.g., DBPedia facts) and related texts (e.g., Wikipedia abstracts) are loosely aligned. We tackle this challenging task by introducing a special-purpose content selection mechanism.1 We use multi-instance learning to automatically discover correspondences between data and text pairs and show how these can be used to enhance the content signal while training an encoder-decoder architecture. Experimental results demonstrate that models trained with content-specific objectives improve upon a vanilla encoder-decoder which solely relies on soft attention.1Our code and data are available at https://github.com/EdinburghNLP/wikigen}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } A core step in statistical data-to-text generation concerns learning correspondences between structured data representations (e.g., facts in a database) and associated texts. In this paper we aim to bootstrap generators from large scale datasets where the data (e.g., DBPedia facts) and related texts (e.g., Wikipedia abstracts) are loosely aligned. We tackle this challenging task by introducing a special-purpose content selection mechanism.1 We use multi-instance learning to automatically discover correspondences between data and text pairs and show how these can be used to enhance the content signal while training an encoder-decoder architecture. Experimental results demonstrate that models trained with content-specific objectives improve upon a vanilla encoder-decoder which solely relies on soft attention.1Our code and data are available at https://github.com/EdinburghNLP/wikigen |
Frermann, Lea; Cohen, Shay; Lapata, M Whodunnit? Crime Drama as a Case for Natural Language Understanding Journal Article Transactions of the Association for Computational Linguistics, 2018, ISSN: 2307-387X. @article{7699a3229d904cfc900ab9683cef2754b, title = {Whodunnit? Crime Drama as a Case for Natural Language Understanding}, author = {Lea Frermann and Shay Cohen and M Lapata}, issn = {2307-387X}, year = {2018}, date = {2018-01-01}, journal = {Transactions of the Association for Computational Linguistics}, abstract = {In this paper we argue that crime drama exemplified in television programs such as CSI: Crime Scene Investigation is an ideal testbed for approximating real-world natural language understanding and the complex inferences associated with it. We propose to treat crime drama as a new inference task, capitalizing on the fact that each episode poses the same basic question (i.e., who committed the crime) and naturally provides the answer when the perpetrator is revealed. We develop a new dataset1 based on CSI episodes, formalize perpetrator identification as a sequence labeling problem, and develop an LSTM-based model which learns from multi-modal data. Experimentalresults show that an incremental inference strategy is key to making accurateguesses as well as learning from representations fusing textual, visual, and acoustic input.}, keywords = {}, pubstate = {published}, tppubtype = {article} } In this paper we argue that crime drama exemplified in television programs such as CSI: Crime Scene Investigation is an ideal testbed for approximating real-world natural language understanding and the complex inferences associated with it. We propose to treat crime drama as a new inference task, capitalizing on the fact that each episode poses the same basic question (i.e., who committed the crime) and naturally provides the answer when the perpetrator is revealed. We develop a new dataset1 based on CSI episodes, formalize perpetrator identification as a sequence labeling problem, and develop an LSTM-based model which learns from multi-modal data. Experimentalresults show that an incremental inference strategy is key to making accurateguesses as well as learning from representations fusing textual, visual, and acoustic input. |
Junczys-Dowmunt, Marcin; Grundkiewicz, Roman; Guha, Shubha; Heafield, Kenneth Approaching Neural Grammatical Error Correction as a Low-Resource Machine Translation Task Inproceedings Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pp. 595–606, Association for Computational Linguistics, 2018. Abstract | Links | BibTeX | Tags: @inproceedings{21393c403acf44e280a3e4b36d43c269, title = {Approaching Neural Grammatical Error Correction as a Low-Resource Machine Translation Task}, author = {Marcin Junczys-Dowmunt and Roman Grundkiewicz and Shubha Guha and Kenneth Heafield}, doi = {10.18653/v1/N18-1055}, year = {2018}, date = {2018-01-01}, booktitle = {Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers)}, pages = {595--606}, publisher = {Association for Computational Linguistics}, abstract = {Previously, neural methods in grammatical error correction (GEC) did not reach state-of-the-art results compared to phrase-based statistical machine translation (SMT) baselines. We demonstrate parallels between neural GEC and low-resource neural MT and successfully adapt several methods from low-resource MT to neural GEC. We further establish guidelines for trustable results in neural GEC and propose a set of model-independent methods for neural GEC that can be easily applied in most GEC settings. Proposed methods include adding source-side noise, domain-adaptation techniques, a GEC-specific training-objective, transfer learning with monolingual data, and ensembling of independently trained GEC models and language models. The combined effects of these methods result in better than state-of-the-art neural GEC models that outperform previously best neural GEC systems by more than 10% M2 on the CoNLL-2014 benchmark and 5.9% on the JFLEG test set. Non-neural state-of-the-art systems are outperformed by more than 2% on the CoNLL-2014benchmark and by 4% on JFLEG.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Previously, neural methods in grammatical error correction (GEC) did not reach state-of-the-art results compared to phrase-based statistical machine translation (SMT) baselines. We demonstrate parallels between neural GEC and low-resource neural MT and successfully adapt several methods from low-resource MT to neural GEC. We further establish guidelines for trustable results in neural GEC and propose a set of model-independent methods for neural GEC that can be easily applied in most GEC settings. Proposed methods include adding source-side noise, domain-adaptation techniques, a GEC-specific training-objective, transfer learning with monolingual data, and ensembling of independently trained GEC models and language models. The combined effects of these methods result in better than state-of-the-art neural GEC models that outperform previously best neural GEC systems by more than 10% M2 on the CoNLL-2014 benchmark and 5.9% on the JFLEG test set. Non-neural state-of-the-art systems are outperformed by more than 2% on the CoNLL-2014benchmark and by 4% on JFLEG. |
Liu, Jiangming; Cohen, Shay; Lapata, Maria Discourse Representation Structure Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, pp. 429–439, Association for Computational Linguistics (ACL), 2018. @inproceedings{96c08df1f6e940dab87c4e35c8cf10bd, title = {Discourse Representation Structure Parsing}, author = {Jiangming Liu and Shay Cohen and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, pages = {429--439}, publisher = {Association for Computational Linguistics (ACL)}, abstract = {We introduce an open-domain neural semantic parser which generates formal meaning representations in the style of Discourse Representation Theory (DRT; Kamp and Reyle 1993). We propose a method which transforms Discourse Representation Structures (DRSs) to trees and develop a structure-aware model which decomposes the decoding process into three stages: basic DRS structure prediction, condition prediction (i.e., predicates and relations), and referent prediction (i.e., variables). Experimental results on the Groningen Meaning Bank (GMB) show that our model outperforms competitive baselines by a wide margin.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We introduce an open-domain neural semantic parser which generates formal meaning representations in the style of Discourse Representation Theory (DRT; Kamp and Reyle 1993). We propose a method which transforms Discourse Representation Structures (DRSs) to trees and develop a structure-aware model which decomposes the decoding process into three stages: basic DRS structure prediction, condition prediction (i.e., predicates and relations), and referent prediction (i.e., variables). Experimental results on the Groningen Meaning Bank (GMB) show that our model outperforms competitive baselines by a wide margin. |
Le, Phong; Titov, Ivan Improving Entity Linking by Modeling Latent Relations between Mentions Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{c4e33d6b2bb9481291fdd6c870068ddf, title = {Improving Entity Linking by Modeling Latent Relations between Mentions}, author = {Phong Le and Ivan Titov}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, abstract = {Entity linking involves aligning textual mentions of named entities to their corresponding entries in a knowledge base. Entity linking systems often exploit relations between textual mentions in a document (e.g., coreference) to decide if the linking decisions are compatible. Unlike previous approaches, which relied on supervised systems or heuristics to predict these relations, we treat relations as latent variables in our neural entity-linking model. We induce the relations without any supervision while optimizing the entity-linking system in an end-to-end fashion. Our multirelational model achieves the best reported scores on the standard benchmark (AIDACoNLL) and substantially outperforms its relation-agnostic version. Its training also converges much faster, suggesting that the injected structural bias helps to explain regularities in the training data.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Entity linking involves aligning textual mentions of named entities to their corresponding entries in a knowledge base. Entity linking systems often exploit relations between textual mentions in a document (e.g., coreference) to decide if the linking decisions are compatible. Unlike previous approaches, which relied on supervised systems or heuristics to predict these relations, we treat relations as latent variables in our neural entity-linking model. We induce the relations without any supervision while optimizing the entity-linking system in an end-to-end fashion. Our multirelational model achieves the best reported scores on the standard benchmark (AIDACoNLL) and substantially outperforms its relation-agnostic version. Its training also converges much faster, suggesting that the injected structural bias helps to explain regularities in the training data. |
Bawden, Rachel; Sennrich, Rico; Birch-Mayne, Alexandra; Haddow, Barry Evaluating Discourse Phenomena in Neural Machine Translation Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. @inproceedings{800395658d054825946113dbefbc3d6a, title = {Evaluating Discourse Phenomena in Neural Machine Translation}, author = {Rachel Bawden and Rico Sennrich and Alexandra Birch-Mayne and Barry Haddow}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {For machine translation to tackle discourse phenomena, models must have access to extrasentential linguistic context. There has been recent interest in modelling context in neural machine translation (NMT), but models have been principally evaluated with standard automatic metrics, poorly adapted to evaluating discourse phenomena. In this article, we present hand-crafted, discourse test sets, designed to test the models’ ability to exploit previous source and target sentences. We investigate the performance of recently proposed multi-encoder NMT models trained on subtitles for English to French. We also explore a novel way of exploiting context from the previous sentence. Despite gains using BLEU, multi-encoder models give limited improvement in the handling of discourse phenomena: 50% accuracy on our coreference test set and 53.5% for coherence/cohesion (compared to a non-contextual baseline of 50%). A simple strategy of decoding the concatenation of the previous and current sentence leads to good performance, and our novel strategy of multiencoding and decoding of two sentences leads to the best performance (72.5% for coreference and 57% for coherence/cohesion), highlighting the importance of target-side context.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } For machine translation to tackle discourse phenomena, models must have access to extrasentential linguistic context. There has been recent interest in modelling context in neural machine translation (NMT), but models have been principally evaluated with standard automatic metrics, poorly adapted to evaluating discourse phenomena. In this article, we present hand-crafted, discourse test sets, designed to test the models’ ability to exploit previous source and target sentences. We investigate the performance of recently proposed multi-encoder NMT models trained on subtitles for English to French. We also explore a novel way of exploiting context from the previous sentence. Despite gains using BLEU, multi-encoder models give limited improvement in the handling of discourse phenomena: 50% accuracy on our coreference test set and 53.5% for coherence/cohesion (compared to a non-contextual baseline of 50%). A simple strategy of decoding the concatenation of the previous and current sentence leads to good performance, and our novel strategy of multiencoding and decoding of two sentences leads to the best performance (72.5% for coreference and 57% for coherence/cohesion), highlighting the importance of target-side context. |
Gella, Spandana; Keller, Frank An Evaluation of Image-based Verb Prediction Models against Human Eye-tracking Data Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 758–763, Association for Computational Linguistics, 2018. Abstract | Links | BibTeX | Tags: @inproceedings{a01f503b085e4544871a81fc342274c6, title = {An Evaluation of Image-based Verb Prediction Models against Human Eye-tracking Data}, author = {Spandana Gella and Frank Keller}, doi = {10.18653/v1/N18-2119}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, pages = {758--763}, publisher = {Association for Computational Linguistics}, abstract = {Recent research in language and vision has developed models for predicting and disambiguating verbs from images. Here, we ask whether the predictions made by such models correspond to human intuitions about visual verbs. We show that the image regions a verb prediction model identifies as salient for a given verb correlate with the regions fixated by human observers performing a verb classification task.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Recent research in language and vision has developed models for predicting and disambiguating verbs from images. Here, we ask whether the predictions made by such models correspond to human intuitions about visual verbs. We show that the image regions a verb prediction model identifies as salient for a given verb correlate with the regions fixated by human observers performing a verb classification task. |
Shoemark, Philippa; Kirby, James; Goldwater, Sharon Inducing a lexicon of sociolinguistic variables from code-mixed text Inproceedings 2018 The 4th Workshop on Noisy User-generated Text (W-NUT), 2018. @inproceedings{3d576ee12d9d4b1b82bf77164c12e0b4, title = {Inducing a lexicon of sociolinguistic variables from code-mixed text}, author = {Philippa Shoemark and James Kirby and Sharon Goldwater}, year = {2018}, date = {2018-01-01}, booktitle = {2018 The 4th Workshop on Noisy User-generated Text (W-NUT)}, abstract = {Sociolinguistics is often concerned with how variants of a linguistic item (e.g., nothing vs. nothin’) are used by different groups or in different situations. We introduce the task of inducing lexical variables from code-mixed text: that is, identifying equivalence pairs such as (football, fitba) along with their linguistic code (football→British, fitba→Scottish). We adapt a framework for identifying gender-biased word pairs to this new task, and present results on three different pairs of English dialects, using tweets as the code-mixed text. Our system achieves precision of over 70% for two of these three datasets, and produces useful results even without extensive parameter tuning. Our success in adapting this framework from gender to language variety suggests that it could be used to discover other types of analogous pairs as well.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Sociolinguistics is often concerned with how variants of a linguistic item (e.g., nothing vs. nothin’) are used by different groups or in different situations. We introduce the task of inducing lexical variables from code-mixed text: that is, identifying equivalence pairs such as (football, fitba) along with their linguistic code (football→British, fitba→Scottish). We adapt a framework for identifying gender-biased word pairs to this new task, and present results on three different pairs of English dialects, using tweets as the code-mixed text. Our system achieves precision of over 70% for two of these three datasets, and produces useful results even without extensive parameter tuning. Our success in adapting this framework from gender to language variety suggests that it could be used to discover other types of analogous pairs as well. |
Louis, Annie; Sutton, Charles Deep Dungeons and Dragons: Learning Character-Action Interactions from Role-Playing Game Transcripts Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. @inproceedings{bf11aa73834349f8934c19bca53fd460, title = {Deep Dungeons and Dragons: Learning Character-Action Interactions from Role-Playing Game Transcripts}, author = {Annie Louis and Charles Sutton}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {An essential aspect to understanding narratives is to grasp the interaction between characters in a story and the actions they take. We examine whether computational models can capture this interaction, when both character attributes and actions are expressed as complex natural language descriptions. We propose role-playing games as a testbed for this problem, and introduce a large corpus1 of game transcripts collected from online discussion forums. Using neural language models which combine character and action descriptions from these stories, we show that we can learn the latent ties. Action sequences are better predicted when the character performing the action is also taken into account, and vice versa for character attributes.1: http://groups.inf.ed.ac.uk/cup/ddd/}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } An essential aspect to understanding narratives is to grasp the interaction between characters in a story and the actions they take. We examine whether computational models can capture this interaction, when both character attributes and actions are expressed as complex natural language descriptions. We propose role-playing games as a testbed for this problem, and introduce a large corpus1 of game transcripts collected from online discussion forums. Using neural language models which combine character and action descriptions from these stories, we show that we can learn the latent ties. Action sequences are better predicted when the character performing the action is also taken into account, and vice versa for character attributes.1: http://groups.inf.ed.ac.uk/cup/ddd/ |
Dong, Li; Quirk, Chris; Lapata, Maria Confidence Modeling for Neural Semantic Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{b2a1033b8ec24df6a77a6d33d66b78de, title = {Confidence Modeling for Neural Semantic Parsing}, author = {Li Dong and Chris Quirk and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, abstract = {In this work we focus on confidence modeling for neural semantic parsers which are built upon sequence-to-sequence models. We outline three major causes of uncertainty, and design various metrics to quantify these factors. These metrics are then used to estimate confidence scores that indicate whether model predictions are likely to be correct. Beyond confidence estimation, we identify which parts of the input contribute to uncertain predictions allowing users to interpret their model, and verify or refine its input. Experimental results show that our confidence model significantly outperforms a widely used method that relies on posterior probability, and improves the quality of interpretation compared to simply relying on attention scores.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } In this work we focus on confidence modeling for neural semantic parsers which are built upon sequence-to-sequence models. We outline three major causes of uncertainty, and design various metrics to quantify these factors. These metrics are then used to estimate confidence scores that indicate whether model predictions are likely to be correct. Beyond confidence estimation, we identify which parts of the input contribute to uncertain predictions allowing users to interpret their model, and verify or refine its input. Experimental results show that our confidence model significantly outperforms a widely used method that relies on posterior probability, and improves the quality of interpretation compared to simply relying on attention scores. |
Grundkiewicz, Roman; Heafield, Kenneth Neural Machine Translation Techniques for Named Entity Transliteration Inproceedings NEWS 2018: The Seventh Named Entities Workshop, 2018. @inproceedings{1b070b5ef95f4136b065195a41ad07dd, title = {Neural Machine Translation Techniques for Named Entity Transliteration}, author = {Roman Grundkiewicz and Kenneth Heafield}, year = {2018}, date = {2018-01-01}, booktitle = {NEWS 2018: The Seventh Named Entities Workshop}, abstract = {Transliterating named entities from one language into another can be approached as neural machine translation (NMT) problem, for which we use deep attentional RNN encoder-decoder models. To build a strong transliteration system, we apply well-established techniques from NMT, such as dropout regularization, model ensembling, rescoring with right-to-left models, and back-translation. Our submission to the NEWS 2018 Shared Task on Named Entity Transliteration ranked first in several tracks}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Transliterating named entities from one language into another can be approached as neural machine translation (NMT) problem, for which we use deep attentional RNN encoder-decoder models. To build a strong transliteration system, we apply well-established techniques from NMT, such as dropout regularization, model ensembling, rescoring with right-to-left models, and back-translation. Our submission to the NEWS 2018 Shared Task on Named Entity Transliteration ranked first in several tracks |
Szubert, Ida; Lopez, Adam; Nathan, Schneider A Structured Syntax-Semantics Interface for English-AMR Alignment Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 1169–1180, Association for Computational Linguistics, 2018. Abstract | Links | BibTeX | Tags: @inproceedings{c0f7fbeb235743c4ad972b565fe7d7a2, title = {A Structured Syntax-Semantics Interface for English-AMR Alignment}, author = {Ida Szubert and Adam Lopez and Schneider Nathan}, doi = {10.18653/v1/N18-1106}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, pages = {1169--1180}, publisher = {Association for Computational Linguistics}, abstract = {Abstract Meaning Representation (AMR) annotations are often assumed to closely mirror dependency syntax, but AMR explicitly does not require this, and the assumption has never been tested. To test it, we devise an expressive framework to align AMR graphs to dependency graphs, which we use to annotate 200 AMRs. Our annotation explains how 97% of AMR edges are evoked by words or syntax. Previously existing AMR alignment frameworks did not allow for mapping AMR onto syntax, and as a consequence they explained at most 23%. While we find that there are indeed many cases where AMR annotations closely mirror syntax, there are also pervasive differences. We use our annotations to test a baseline AMR-to-syntax aligner, finding that this task is more difficult than AMRto-string alignment; and to pinpoint errors in an AMR parser. We make our data and code freely available for further research on AMR parsing and generation, and the relationship of AMR to syntax.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Abstract Meaning Representation (AMR) annotations are often assumed to closely mirror dependency syntax, but AMR explicitly does not require this, and the assumption has never been tested. To test it, we devise an expressive framework to align AMR graphs to dependency graphs, which we use to annotate 200 AMRs. Our annotation explains how 97% of AMR edges are evoked by words or syntax. Previously existing AMR alignment frameworks did not allow for mapping AMR onto syntax, and as a consequence they explained at most 23%. While we find that there are indeed many cases where AMR annotations closely mirror syntax, there are also pervasive differences. We use our annotations to test a baseline AMR-to-syntax aligner, finding that this task is more difficult than AMRto-string alignment; and to pinpoint errors in an AMR parser. We make our data and code freely available for further research on AMR parsing and generation, and the relationship of AMR to syntax. |
Philip John, ; Lapata, Maria What’s This Movie About? A Joint Neural Network Architecture for Movie Content Analysis Inproceedings Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 1770–1781, Association for Computational Linguistics (ACL), 2018. Abstract | Links | BibTeX | Tags: Information Extraction, Natural Language Generation, Natural Language Processing, Neural Networks, Summarisation @inproceedings{a538dbbfb5e343f5a0dcf1e8d057ddb1, title = {What’s This Movie About? A Joint Neural Network Architecture for Movie Content Analysis}, author = {{Philip John} Gorinski and Maria Lapata}, doi = {10.18653/v1/N18-1160}, year = {2018}, date = {2018-01-01}, booktitle = {Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, volume = {1}, pages = {1770--1781}, publisher = {Association for Computational Linguistics (ACL)}, abstract = {This work takes a first step toward movie content analysis by tackling the novel task of movie overview generation. Overviews are natural language texts that give a first impression of a movie, describing aspects such as its genre, plot, mood, or artistic style. We create a dataset that consists of movie scripts, attribute-value pairs for the movies' aspects, as well as overviews, which we extract from an online database. We present a novel end-to-end model for overview generation, consisting of a multi-label encoder for identifying screenplay attributes, and an LSTM decoder to generate natural language sentences conditioned on the identified attributes. Automatic and human evaluation show that the encoder is able to reliably assign good labels for the movie's attributes, and the overviews provide descriptions of the movie's content which are informative and faithful.}, keywords = {Information Extraction, Natural Language Generation, Natural Language Processing, Neural Networks, Summarisation}, pubstate = {published}, tppubtype = {inproceedings} } This work takes a first step toward movie content analysis by tackling the novel task of movie overview generation. Overviews are natural language texts that give a first impression of a movie, describing aspects such as its genre, plot, mood, or artistic style. We create a dataset that consists of movie scripts, attribute-value pairs for the movies' aspects, as well as overviews, which we extract from an online database. We present a novel end-to-end model for overview generation, consisting of a multi-label encoder for identifying screenplay attributes, and an LSTM decoder to generate natural language sentences conditioned on the identified attributes. Automatic and human evaluation show that the encoder is able to reliably assign good labels for the movie's attributes, and the overviews provide descriptions of the movie's content which are informative and faithful. |
Grover, Claire; Tobin, Richard; Alex, Beatrice; Sudlow, Catherine; Mair, Grant; Whiteley, William Text Mining Brain Imaging Reports Conference 2018, (UK Healthcare Text Analytics Conference (HealTAC-2018) : Unlocking Evidence Contained in Healthcare Free-text, HealTAC-2018 ; Conference date: 18-04-2018 Through 19-04-2018). Abstract | Links | BibTeX | Tags: @conference{252f1fd829804ea9a54de75758eb1589, title = {Text Mining Brain Imaging Reports}, author = {Claire Grover and Richard Tobin and Beatrice Alex and Catherine Sudlow and Grant Mair and William Whiteley}, url = {http://healtex.org/healtac-2018/}, year = {2018}, date = {2018-01-01}, abstract = {We describe a text mining system for classifying radiologists’ reports of CT and MRI brain scans, assigning labels indicating occurrence and type of stroke, as well as other observations. Our system, the Edinburgh Information Extraction for Radiology reports (EdIE-R) system, was developed and tested on a collection of 1,168 reports from the Edinburgh Stroke Study (ESS), a hospital-based register of stroke and transient ischaemic attack patients. Automated reading of reports such as these opens up avenues for population health monitoring and audit, and can provide a resource for epidemiological studies. Here we describe the EdIE-R system and report on its development and evaluation on annotated data from ESS. We aim to make the development and testing annotations for the ESS collection available for research.}, note = {UK Healthcare Text Analytics Conference (HealTAC-2018) : Unlocking Evidence Contained in Healthcare Free-text, HealTAC-2018 ; Conference date: 18-04-2018 Through 19-04-2018}, keywords = {}, pubstate = {published}, tppubtype = {conference} } We describe a text mining system for classifying radiologists’ reports of CT and MRI brain scans, assigning labels indicating occurrence and type of stroke, as well as other observations. Our system, the Edinburgh Information Extraction for Radiology reports (EdIE-R) system, was developed and tested on a collection of 1,168 reports from the Edinburgh Stroke Study (ESS), a hospital-based register of stroke and transient ischaemic attack patients. Automated reading of reports such as these opens up avenues for population health monitoring and audit, and can provide a resource for epidemiological studies. Here we describe the EdIE-R system and report on its development and evaluation on annotated data from ESS. We aim to make the development and testing annotations for the ESS collection available for research. |
Cremarenco, Diana; Magdy, Walid ClassStrength v2: An Adaptive Multilingual Tool for Tweet Classification Inproceedings 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), 2018. @inproceedings{2ac409d9fbfe48a79f6a5e4d673e4ade, title = {ClassStrength v2: An Adaptive Multilingual Tool for Tweet Classification}, author = {Diana Cremarenco and Walid Magdy}, year = {2018}, date = {2018-01-01}, booktitle = {2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM)}, abstract = {In this paper we present the second version of our multilingual tweet classification tool. ClassStrength v2 classifies tweets into 14 categories (Sports, Music, News&Politics etc.) using a distant supervision approach. The new version extends the initial set of five languages to ten (English, French, German, Chinese, Japanese, Arabic, Russian, Spanish, Portuguese and Polish). In addition, the classification models of each language get automatically updated every month to allow accurate classification over time. Our experimentation showed that the larger the time gap between the tweet and the data used for training the model, the worse the performance, which motivated for creating an adaptive version of ClassStrength that get its models updated periodically.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } In this paper we present the second version of our multilingual tweet classification tool. ClassStrength v2 classifies tweets into 14 categories (Sports, Music, News&Politics etc.) using a distant supervision approach. The new version extends the initial set of five languages to ten (English, French, German, Chinese, Japanese, Arabic, Russian, Spanish, Portuguese and Polish). In addition, the classification models of each language get automatically updated every month to allow accurate classification over time. Our experimentation showed that the larger the time gap between the tweet and the data used for training the model, the worse the performance, which motivated for creating an adaptive version of ClassStrength that get its models updated periodically. |
Issa, Fuad; Damonte, Marco; Cohen, Shay; Yan, Xiaohui; Chang, Yi Abstract Meaning Representation for Paraphrase Detection Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 442–452, Association for Computational Linguistics, 2018. Abstract | Links | BibTeX | Tags: @inproceedings{6b132e2c1b80423b93783cc7959fcb29, title = {Abstract Meaning Representation for Paraphrase Detection}, author = {Fuad Issa and Marco Damonte and Shay Cohen and Xiaohui Yan and Yi Chang}, doi = {10.18653/v1/N18-1041}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, pages = {442--452}, publisher = {Association for Computational Linguistics}, abstract = {Abstract Meaning Representation (AMR) parsing aims at abstracting away from the syntactic realization of a sentence, and denoting only its meaning in a canonical form. As such, it is ideal for paraphrase detection, a problem in which one is required to specify whether two sentences have the same meaning. We show that na"ive use of AMR in paraphrase detection is not necessarily useful, and turn to describe a technique based on latent semantic analysis in combination with AMR parsing that significantly advances state-of-the-art results in paraphrase detection for the Microsoft Research Paraphrase Corpus. Our best results in the transductive setting are 86.6% for accuracy and 90.0% for F1 measure.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Abstract Meaning Representation (AMR) parsing aims at abstracting away from the syntactic realization of a sentence, and denoting only its meaning in a canonical form. As such, it is ideal for paraphrase detection, a problem in which one is required to specify whether two sentences have the same meaning. We show that na"ive use of AMR in paraphrase detection is not necessarily useful, and turn to describe a technique based on latent semantic analysis in combination with AMR parsing that significantly advances state-of-the-art results in paraphrase detection for the Microsoft Research Paraphrase Corpus. Our best results in the transductive setting are 86.6% for accuracy and 90.0% for F1 measure. |
Junczys-Dowmunt, Marcin; Grundkiewicz, Roman; Dwojak, Tomasz; Hoang, Hieu; Heafield, Kenneth; Neckermann, Tom; Seide, Frank; Germann, Ulrich; Alham Fikri, ; Bogoychev, Nikolay; André F T, ; Birch-Mayne, Alexandra Marian: Fast Neural Machine Translation in C++ Inproceedings The 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{6a9a69b68b664ca7bdf24a8f65f30ac5, title = {Marian: Fast Neural Machine Translation in C++}, author = {Marcin Junczys-Dowmunt and Roman Grundkiewicz and Tomasz Dwojak and Hieu Hoang and Kenneth Heafield and Tom Neckermann and Frank Seide and Ulrich Germann and {Alham Fikri} Aji and Nikolay Bogoychev and {André F T } Martins and Alexandra Birch-Mayne}, year = {2018}, date = {2018-01-01}, booktitle = {The 56th Annual Meeting of the Association for Computational Linguistics}, abstract = {We present Marian, an efficient and self-contained Neural Machine Translation framework with an integrated automatic differentiation engine based on dynamic computation graphs. Marian is written entirely in C++. We describe the design of the encoder-decoder framework and demonstrate that a research-friendly toolkit can achieve high training and translation speed.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We present Marian, an efficient and self-contained Neural Machine Translation framework with an integrated automatic differentiation engine based on dynamic computation graphs. Marian is written entirely in C++. We describe the design of the encoder-decoder framework and demonstrate that a research-friendly toolkit can achieve high training and translation speed. |
Liu, Yang; Gardner, Matt; Lapata, Maria Structured Alignment Networks for Matching Sentences Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. @inproceedings{9cb6b9dd80724113a1d05c6b492e5286, title = {Structured Alignment Networks for Matching Sentences}, author = {Yang Liu and Matt Gardner and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {2018 Conference on Empirical Methods in Natural Language Processing}, abstract = {Many tasks in natural language processing involve comparing two sentences to compute some notion of relevance, entailment, or similarity. Typically, this comparison is done either at the word level or at the sentence level, with no attempt to leverage the inherent structure of the sentence. When sentence structure is used for comparison, it is obtained during a non-differentiable pre-processing step, leading to propagation of errors. We introduce a model of structured alignments between sentences, showing how to compare two sentences by matching their latent structures. Using a structured attention mechanism, our model matches candidate spans in the first sentence to candidate spans in the second sentence, simultaneously discovering the tree structure of each sentence. Our model is fully differentiable and trained only on the matching objective. We evaluate this model on two tasks, natural entailment detection and answer sentence selection, and find that modeling latent tree structures results in superior performance. Analysis of the learned sentence structures shows they can reflect some syntactic phenomena.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Many tasks in natural language processing involve comparing two sentences to compute some notion of relevance, entailment, or similarity. Typically, this comparison is done either at the word level or at the sentence level, with no attempt to leverage the inherent structure of the sentence. When sentence structure is used for comparison, it is obtained during a non-differentiable pre-processing step, leading to propagation of errors. We introduce a model of structured alignments between sentences, showing how to compare two sentences by matching their latent structures. Using a structured attention mechanism, our model matches candidate spans in the first sentence to candidate spans in the second sentence, simultaneously discovering the tree structure of each sentence. Our model is fully differentiable and trained only on the matching objective. We evaluate this model on two tasks, natural entailment detection and answer sentence selection, and find that modeling latent tree structures results in superior performance. Analysis of the learned sentence structures shows they can reflect some syntactic phenomena. |
Brazinskas, Arthur; Havrylov, Serhii; Titov, Ivan Embedding Words as Distributions with a Bayesian Skip-gram Model Inproceedings 27th International Conference on Computational Linguistics, 2018. @inproceedings{74dc811afe31489dbe2dbad8f05e36b0, title = {Embedding Words as Distributions with a Bayesian Skip-gram Model}, author = {Arthur Brazinskas and Serhii Havrylov and Ivan Titov}, year = {2018}, date = {2018-01-01}, booktitle = {27th International Conference on Computational Linguistics}, abstract = {We introduce a method for embedding words as probability densities in a low-dimensional space. Rather than assuming that a word embedding is fixed across the entire text collection, as in standard word embedding methods, in our Bayesian model we generate it from a word-specific prior density for each occurrence of a given word. Intuitively, for each word, the prior density encodes the distribution of its potential ‘meanings’. These prior densities are conceptually similar to Gaussian embeddings of Vilnis and McCallum (2015). Interestingly, unlike the Gaussian embeddings, we can also obtain context-specific densities: they encode uncertainty about the sense of a word given its context and correspond to the approximate posterior distributions within our model. The context-dependent densities have many potential applications: for example, we show that they can be directly used in the lexical substitution task. We describe an effective estimation method based on the variational autoencoding framework and demonstrate the effectiveness of our embedding technique on a range of standard benchmarks.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We introduce a method for embedding words as probability densities in a low-dimensional space. Rather than assuming that a word embedding is fixed across the entire text collection, as in standard word embedding methods, in our Bayesian model we generate it from a word-specific prior density for each occurrence of a given word. Intuitively, for each word, the prior density encodes the distribution of its potential ‘meanings’. These prior densities are conceptually similar to Gaussian embeddings of Vilnis and McCallum (2015). Interestingly, unlike the Gaussian embeddings, we can also obtain context-specific densities: they encode uncertainty about the sense of a word given its context and correspond to the approximate posterior distributions within our model. The context-dependent densities have many potential applications: for example, we show that they can be directly used in the lexical substitution task. We describe an effective estimation method based on the variational autoencoding framework and demonstrate the effectiveness of our embedding technique on a range of standard benchmarks. |
Papasarantopoulos, Nikolaos; Jiang, Helen; Cohen, Shay Canonical Correlation Inference for Mapping Abstract Scenes to Text Inproceedings Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18), 2018. @inproceedings{14a893dc9c8b4363b05b8c50f365bfc2b, title = {Canonical Correlation Inference for Mapping Abstract Scenes to Text}, author = {Nikolaos Papasarantopoulos and Helen Jiang and Shay Cohen}, year = {2018}, date = {2018-01-01}, booktitle = {Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)}, abstract = {We describe a technique for structured prediction, based on canonical correlation analysis. Our learning algorithm finds two projections for the input and the output spaces that aim at projecting a given input and its correct output into points close to each other. We demonstrate our technique on a language-vision problem, namely the problem of giving a textual description to an “abstract scene.”}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We describe a technique for structured prediction, based on canonical correlation analysis. Our learning algorithm finds two projections for the input and the output spaces that aim at projecting a given input and its correct output into points close to each other. We demonstrate our technique on a language-vision problem, namely the problem of giving a textual description to an “abstract scene.” |
Frank, Stella; Elliott, Desmond; Specia, Lucia Assessing multilingual multimodal image description: Studies of native speaker preferences and translator choices Journal Article Natural Language Engineering, 24 (3), pp. 393–413, 2018, ISSN: 1351-3249, (Route 1). Abstract | Links | BibTeX | Tags: @article{e4f4a3fa2d774c0abe72dead55bf3780b, title = {Assessing multilingual multimodal image description: Studies of native speaker preferences and translator choices}, author = {Stella Frank and Desmond Elliott and Lucia Specia}, doi = {10.1017/S1351324918000074}, issn = {1351-3249}, year = {2018}, date = {2018-01-01}, journal = {Natural Language Engineering}, volume = {24}, number = {3}, pages = {393--413}, publisher = {Cambridge University Press}, abstract = {Two studies on multilingual multimodal image description provide empirical evidence towards two hypotheses at the core of the task: (i) whether target language speakers pre- fer descriptions generated directly in their native language, as compared to descriptions translated from a different language; (ii) the role of the image in human translation of de- scriptions. These results provide guidance for future work in multimodal natural language processing by firstly showing that on the whole, translations are not distinguished from native language descriptions, and secondly delineating and quantifying the information gained from the image during the human translation task.}, note = {Route 1}, keywords = {}, pubstate = {published}, tppubtype = {article} } Two studies on multilingual multimodal image description provide empirical evidence towards two hypotheses at the core of the task: (i) whether target language speakers pre- fer descriptions generated directly in their native language, as compared to descriptions translated from a different language; (ii) the role of the image in human translation of de- scriptions. These results provide guidance for future work in multimodal natural language processing by firstly showing that on the whole, translations are not distinguished from native language descriptions, and secondly delineating and quantifying the information gained from the image during the human translation task. |
Elsayed, Tamer; Magdy, Walid; Kutlu, Mucahid; Hasanain, Maram; Suwaileh, Reem BroDyn’18: Workshop on analysis of broad dynamic topics over social media Inproceedings Advances in Information Retrieval - 40th European Conference on IR Research, ECIR 2018, Proceedings, pp. 829–830, Springer-Verlag, 2018, ISBN: 978-3-319-76940-0. @inproceedings{3f8a26fb5113401f802a2d1d363daa7a, title = {BroDyn’18: Workshop on analysis of broad dynamic topics over social media}, author = {Tamer Elsayed and Walid Magdy and Mucahid Kutlu and Maram Hasanain and Reem Suwaileh}, doi = {10.1007/978-3-319-76941-7}, isbn = {978-3-319-76940-0}, year = {2018}, date = {2018-01-01}, booktitle = {Advances in Information Retrieval - 40th European Conference on IR Research, ECIR 2018, Proceedings}, volume = {10772 LNCS}, pages = {829--830}, publisher = {Springer-Verlag}, series = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
Narayan, Shashi; Cardenas, Ronald; Papasarantopoulos, Nikolaos; Cohen, Shay; Lapata, Mirella; Yu, Jiangsheng; Chang, Yi Document Modeling with External Attention for Sentence Extraction Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{a56f3ea850a84af4a8f11ebf31a68f69, title = {Document Modeling with External Attention for Sentence Extraction}, author = {Shashi Narayan and Ronald Cardenas and Nikolaos Papasarantopoulos and Shay Cohen and Mirella Lapata and Jiangsheng Yu and Yi Chang}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, abstract = {Document modeling is essential to a variety of natural language understanding tasks. We propose to use external information to improve document modeling for problems that can be framed as sentence extraction. We develop a framework composed of a hierarchical document encoder and an attention-based extractor with attention over external information. We evaluate our model on extractive document summarization (where the external information is image captions and the title of the document) and answer selection (where the external information is a question). We show that our model consistently outperforms strong baselines, in terms of both informativeness and fluency (for CNN document summarization) and achieves state-of-the-art results for answer selection on WikiQA and NewsQA.11Our TensorFlow code and datasets are publicly available at https://github.com/shashiongithub/Document-Models-with-Ext Information. }, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Document modeling is essential to a variety of natural language understanding tasks. We propose to use external information to improve document modeling for problems that can be framed as sentence extraction. We develop a framework composed of a hierarchical document encoder and an attention-based extractor with attention over external information. We evaluate our model on extractive document summarization (where the external information is image captions and the title of the document) and answer selection (where the external information is a question). We show that our model consistently outperforms strong baselines, in terms of both informativeness and fluency (for CNN document summarization) and achieves state-of-the-art results for answer selection on WikiQA and NewsQA.11Our TensorFlow code and datasets are publicly available at https://github.com/shashiongithub/Document-Models-with-Ext Information. |
Schlichtkrull, Michael; Thomas N, ; Bloem, Peter; van den Berg, Rianne ; Titov, Ivan; Welling, Max Modeling Relational Data with Graph Convolutional Networks Inproceedings The 15th Extended Semantic Web Conference (ESWC), 2018. @inproceedings{e5106e47402e472894b0b528b1eb0985, title = {Modeling Relational Data with Graph Convolutional Networks}, author = {Michael Schlichtkrull and {Thomas N } Kipf and Peter Bloem and Rianne {van den Berg} and Ivan Titov and Max Welling}, year = {2018}, date = {2018-01-01}, booktitle = {The 15th Extended Semantic Web Conference (ESWC)}, abstract = {Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8% on FB15k-237 over a decoder-only baseline.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Knowledge graphs enable a wide variety of applications, including question answering and information retrieval. Despite the great effort invested in their creation and maintenance, even the largest (e.g., Yago, DBPedia or Wikidata) remain incomplete. We introduce Relational Graph Convolutional Networks (R-GCNs) and apply them to two standard knowledge base completion tasks: Link prediction (recovery of missing facts, i.e. subject-predicate-object triples) and entity classification (recovery of missing entity attributes). R-GCNs are related to a recent class of neural networks operating on graphs, and are developed specifically to handle the highly multi-relational data characteristic of realistic knowledge bases. We demonstrate the effectiveness of R-GCNs as a stand-alone model for entity classification. We further show that factorization models for link prediction such as DistMult can be significantly improved through the use of an R-GCN encoder model to accumulate evidence over multiple inference steps in the graph, demonstrating a large improvement of 29.8% on FB15k-237 over a decoder-only baseline. |
Voita, Elena; Serdyukov, Pavel; Sennrich, Rico; Titov, Ivan Context-Aware Neural Machine Translation Learns Anaphora Resolution Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{8b04c5cb3e56494186a274ee3292d427, title = {Context-Aware Neural Machine Translation Learns Anaphora Resolution}, author = {Elena Voita and Pavel Serdyukov and Rico Sennrich and Ivan Titov}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, abstract = {Standard machine translation systems process sentences in isolation and hence ignore extra-sentential information, even though extended context can both prevent mistakes in ambiguous cases and improve translation coherence. We introduce a context-aware neural machine translation model designed in such way that the flow of information from the extended context to the translation model can be controlled and analyzed. We experiment with an English-Russian subtitles dataset, and observe that much of what is captured by our model deals with improving pronoun translation. We measure correspondences between induced attention distributions and coreference relations and observe that the model implicitly captures anaphora. It is consistent with gains for sentences where pronouns need to be gendered in translation. Beside improvements in anaphoric cases, the model also improves in overall BLEU, both over its context-agnostic version (+0.7) and over simple concatenation of the context and source sentences (+0.6).}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Standard machine translation systems process sentences in isolation and hence ignore extra-sentential information, even though extended context can both prevent mistakes in ambiguous cases and improve translation coherence. We introduce a context-aware neural machine translation model designed in such way that the flow of information from the extended context to the translation model can be controlled and analyzed. We experiment with an English-Russian subtitles dataset, and observe that much of what is captured by our model deals with improving pronoun translation. We measure correspondences between induced attention distributions and coreference relations and observe that the model implicitly captures anaphora. It is consistent with gains for sentences where pronouns need to be gendered in translation. Beside improvements in anaphoric cases, the model also improves in overall BLEU, both over its context-agnostic version (+0.7) and over simple concatenation of the context and source sentences (+0.6). |
Alharbi, Randah; Magdy, Walid; Darwish, Kareem; AbdelAli, Ahmed; Mubarak, Hamdy Part-of-Speech Tagging for Arabic Gulf Dialect Using Bi-LSTM Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. Abstract | BibTeX | Tags: arabic, BiLSTM, POS @inproceedings{f7643c7c28d24150b99450cc18001c73b, title = {Part-of-Speech Tagging for Arabic Gulf Dialect Using Bi-LSTM}, author = {Randah Alharbi and Walid Magdy and Kareem Darwish and Ahmed AbdelAli and Hamdy Mubarak}, year = {2018}, date = {2018-01-01}, booktitle = {11th edition of the Language Resources and Evaluation Conference}, abstract = {Part-of-speech (POS) tagging is one of the most important addressed areas in the natural language processing (NLP). There are effective POS taggers for many languages including Arabic. However, POS research for Arabic focused mainly on Modern Standard Arabic (MSA), while less attention was directed towards Dialect Arabic (DA). MSA is the formal variant which is mainly found in news and formal text books, while DA is the informal spoken Arabic that varies among different regions in the Arab world. DA is heavily used online due to the large spread of social media, which increased research directions towards building NLP tools for DA. Most research on DA focuses on Egyptian and Levantine, while much less attention is given to the Gulf dialect. In this paper, we present a more effective POS tagger for the Arabic Gulf dialect than currently available Arabic POS taggers. Our work includes preparing a POS tagging dataset, engineering multiple sets of features, and applying two machine learning methods, namely Support Vector Machine (SVM) classifier and bi-directional Long Short Term Memory (Bi-LSTM) for sequence modeling. We have improved POS tagging for Gulf dialect from 75% accuracy using a state-of-the-art MSA POS tagger to over 91% accuracy using a Bi-LSTM labeler.}, keywords = {arabic, BiLSTM, POS}, pubstate = {published}, tppubtype = {inproceedings} } Part-of-speech (POS) tagging is one of the most important addressed areas in the natural language processing (NLP). There are effective POS taggers for many languages including Arabic. However, POS research for Arabic focused mainly on Modern Standard Arabic (MSA), while less attention was directed towards Dialect Arabic (DA). MSA is the formal variant which is mainly found in news and formal text books, while DA is the informal spoken Arabic that varies among different regions in the Arab world. DA is heavily used online due to the large spread of social media, which increased research directions towards building NLP tools for DA. Most research on DA focuses on Egyptian and Levantine, while much less attention is given to the Gulf dialect. In this paper, we present a more effective POS tagger for the Arabic Gulf dialect than currently available Arabic POS taggers. Our work includes preparing a POS tagging dataset, engineering multiple sets of features, and applying two machine learning methods, namely Support Vector Machine (SVM) classifier and bi-directional Long Short Term Memory (Bi-LSTM) for sequence modeling. We have improved POS tagging for Gulf dialect from 75% accuracy using a state-of-the-art MSA POS tagger to over 91% accuracy using a Bi-LSTM labeler. |
Narayan, Shashi; Cohen, Shay; Lapata, Maria Ranking Sentences for Extractive Summarization with Reinforcement Learning Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. @inproceedings{7cf77471c6b04e2687310f785b811ad0, title = {Ranking Sentences for Extractive Summarization with Reinforcement Learning}, author = {Shashi Narayan and Shay Cohen and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {Single document summarization is the task of producing a shorter version of a document while preserving its principal information content. In this paper we conceptualize extractive summarization as a sentence ranking task and propose a novel training algorithm which globally optimizes the ROUGE evaluation metric through a reinforcement learning objective. We use our algorithm to train a neural summarization model on the CNN and DailyMail datasets and demonstrate experimentally that it outperforms state-of-the-art extractive and abstractive systems when evaluated automatically and by humans.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Single document summarization is the task of producing a shorter version of a document while preserving its principal information content. In this paper we conceptualize extractive summarization as a sentence ranking task and propose a novel training algorithm which globally optimizes the ROUGE evaluation metric through a reinforcement learning objective. We use our algorithm to train a neural summarization model on the CNN and DailyMail datasets and demonstrate experimentally that it outperforms state-of-the-art extractive and abstractive systems when evaluated automatically and by humans. |
Angelidis, Stefanos; Lapata, Maria Multiple Instance Learning Networks for Fine-Grained Sentiment Analysis Journal Article Transactions of the Association for Computational Linguistics, 6 , pp. 17–32, 2018, ISSN: 2307-387X. @article{9b1057529eb84c598cad2fbc13702c8d, title = {Multiple Instance Learning Networks for Fine-Grained Sentiment Analysis}, author = {Stefanos Angelidis and Maria Lapata}, issn = {2307-387X}, year = {2018}, date = {2018-01-01}, journal = {Transactions of the Association for Computational Linguistics}, volume = {6}, pages = {17--32}, abstract = {We consider the task of fine-grained sentiment analysis from the perspective of multiple instance learning (MIL). Our neural model is trained on document sentiment labels, and learns to predict the sentiment of text segments, i.e. sentences or elementary discourse units (EDUs), without segment-level supervision. We introduce an attention-based polarity scoring method for identifying positive and negative text snippets and a new dataset which we call SPOT (as shorthand for Segment-level POlariTy annotations) for evaluating MIL-style sentiment models like ours. Experimental results demonstrate superior performance against multiple baselines, whereas a judgement elicitation study shows that EDU-level opinion extraction produces more informative summaries than sentence-based alternatives.}, keywords = {}, pubstate = {published}, tppubtype = {article} } We consider the task of fine-grained sentiment analysis from the perspective of multiple instance learning (MIL). Our neural model is trained on document sentiment labels, and learns to predict the sentiment of text segments, i.e. sentences or elementary discourse units (EDUs), without segment-level supervision. We introduce an attention-based polarity scoring method for identifying positive and negative text snippets and a new dataset which we call SPOT (as shorthand for Segment-level POlariTy annotations) for evaluating MIL-style sentiment models like ours. Experimental results demonstrate superior performance against multiple baselines, whereas a judgement elicitation study shows that EDU-level opinion extraction produces more informative summaries than sentence-based alternatives. |
Zhang, Xingxing; Lapata, Maria; Wei, Furu; Zhou, Ming Neural Latent Extractive Document Summarization Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. @inproceedings{25ad917237c84817a67a8c6ac9cd0296, title = {Neural Latent Extractive Document Summarization}, author = {Xingxing Zhang and Maria Lapata and Furu Wei and Ming Zhou}, year = {2018}, date = {2018-01-01}, booktitle = {2018 Conference on Empirical Methods in Natural Language Processing}, abstract = {Extractive summarization models require sentence-level labels, which are usually created heuristically (e.g., with rule-based methods) given that most summarization datasets only have document-summary pairs. Since these labels might be suboptimal, we propose a latent variable extractive model where sentences are viewed as latent variables and sentences with activated variables are used to infer gold summaries. During training the loss comes directly from gold summaries. Experiments on the CNN/Dailymail dataset show that our model improves over a strong extractive baseline trained on heuristically approximated labels and also performs competitively to several recent models.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Extractive summarization models require sentence-level labels, which are usually created heuristically (e.g., with rule-based methods) given that most summarization datasets only have document-summary pairs. Since these labels might be suboptimal, we propose a latent variable extractive model where sentences are viewed as latent variables and sentences with activated variables are used to infer gold summaries. During training the loss comes directly from gold summaries. Experiments on the CNN/Dailymail dataset show that our model improves over a strong extractive baseline trained on heuristically approximated labels and also performs competitively to several recent models. |
Robertson, Alexander; Magdy, Walid; Goldwater, Sharon Self-Representation on Twitter Using Emoji Skin Color Modifiers Inproceedings 12th International AAAI Conference on Web and Social Media (ICWSM-18), 2018. @inproceedings{f64e209e76df46e98b299ace96a00d83, title = {Self-Representation on Twitter Using Emoji Skin Color Modifiers}, author = {Alexander Robertson and Walid Magdy and Sharon Goldwater}, year = {2018}, date = {2018-01-01}, booktitle = {12th International AAAI Conference on Web and Social Media (ICWSM-18)}, abstract = {Since 2015, it has been possible to modify certain emoji with a skin tone. The five different skin tones were introduced with the aim of representing more human diversity, but some commentators feared they might be used as a way to negatively represent other users/groups. This paper presents a quantitative analysis of the use of skin tone modifiers on emoji on Twitter, showing that users with darker-skinned profile photos employ them more often than users with lighter-skinned profile photos, and the vast majority of skin tone usage matches the color of a user’s profile photo—i.e., tones represent the self, rather than the other. In the few cases where users do use opposite-toned emoji, we find no evidence of negative racial sentiment. Thus, the introduction of skin tones seems to have met the goal of better representing human diversity.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Since 2015, it has been possible to modify certain emoji with a skin tone. The five different skin tones were introduced with the aim of representing more human diversity, but some commentators feared they might be used as a way to negatively represent other users/groups. This paper presents a quantitative analysis of the use of skin tone modifiers on emoji on Twitter, showing that users with darker-skinned profile photos employ them more often than users with lighter-skinned profile photos, and the vast majority of skin tone usage matches the color of a user’s profile photo—i.e., tones represent the self, rather than the other. In the few cases where users do use opposite-toned emoji, we find no evidence of negative racial sentiment. Thus, the introduction of skin tones seems to have met the goal of better representing human diversity. |
Robertson, Alexander; Goldwater, Sharon Evaluating Historical Text Normalization Systems: How Well Do They Generalize? Inproceedings 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. @inproceedings{7684173ef1e8456783818d8488b0dd1c, title = {Evaluating Historical Text Normalization Systems: How Well Do They Generalize?}, author = {Alexander Robertson and Sharon Goldwater}, year = {2018}, date = {2018-01-01}, booktitle = {16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {We highlight several issues in the evaluation of historical text normalization systems that make it hard to tell how well these systems would actually work in practice—i.e., for new datasets or languages; in comparison to more na"ive systems; or as a preprocessing step for downstream NLP tools. We illustrate these issues and exemplify our proposed evaluation practices by comparing two neural models against a na"ive baseline system. We show that the neural models generalize well to unseen words in tests on five languages; nevertheless, they provide no clear benefit over the na"ive baseline for downstream POS tagging of an English historical collection. We conclude that future work should include more rigorous evaluation, including both intrinsic and extrinsic measures where possible.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We highlight several issues in the evaluation of historical text normalization systems that make it hard to tell how well these systems would actually work in practice—i.e., for new datasets or languages; in comparison to more na"ive systems; or as a preprocessing step for downstream NLP tools. We illustrate these issues and exemplify our proposed evaluation practices by comparing two neural models against a na"ive baseline system. We show that the neural models generalize well to unseen words in tests on five languages; nevertheless, they provide no clear benefit over the na"ive baseline for downstream POS tagging of an English historical collection. We conclude that future work should include more rigorous evaluation, including both intrinsic and extrinsic measures where possible. |
Angelidis, Stefanos; Lapata, Maria Summarizing Opinions: Aspect Extraction Meets Sentiment Prediction and They Are Both Weakly Supervised Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. @inproceedings{618b503e00a04074b44128c5884b8fda, title = {Summarizing Opinions: Aspect Extraction Meets Sentiment Prediction and They Are Both Weakly Supervised}, author = {Stefanos Angelidis and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {2018 Conference on Empirical Methods in Natural Language Processing}, abstract = {We present a neural framework for opinion summarization from online product reviews which is knowledge-lean and only requires light supervision (e.g., in the form of product domain labels and user-provided ratings). Our method combines two weakly supervised components to identify salient opinions and form extractive summaries from multiple reviews: an aspect extractor trained under a multi-task objective, and a sentiment predictor based on multiple instance learning. We introduce an opinion summarization dataset that includes a training set of product reviews from six diverse domains and human-annotated development and test sets with gold standard aspect annotations, salience labels, and opinion summaries. Automatic evaluation shows significant improvements over baselines, and a largescale study indicates that our opinion summaries are preferred by human judges according to multiple criteria.1}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We present a neural framework for opinion summarization from online product reviews which is knowledge-lean and only requires light supervision (e.g., in the form of product domain labels and user-provided ratings). Our method combines two weakly supervised components to identify salient opinions and form extractive summaries from multiple reviews: an aspect extractor trained under a multi-task objective, and a sentiment predictor based on multiple instance learning. We introduce an opinion summarization dataset that includes a training set of product reviews from six diverse domains and human-annotated development and test sets with gold standard aspect annotations, salience labels, and opinion summaries. Automatic evaluation shows significant improvements over baselines, and a largescale study indicates that our opinion summaries are preferred by human judges according to multiple criteria.1 |
Hoang, Hieu; Dwojak, Tomasz; Krislauks, Rihards; Torregrosa, Daniel; Heafield, Kenneth Fast Neural Machine Translation Implementation Inproceedings The 2nd Workshop on Neural Machine Translation and Generation 2018, 2018. @inproceedings{060f6ea9274141ffb21b73bfdc81c0f5, title = {Fast Neural Machine Translation Implementation}, author = {Hieu Hoang and Tomasz Dwojak and Rihards Krislauks and Daniel Torregrosa and Kenneth Heafield}, year = {2018}, date = {2018-01-01}, booktitle = {The 2nd Workshop on Neural Machine Translation and Generation 2018}, abstract = {This paper describes the submissions to the efficiency track for GPUs by members of the University of Edinburgh, Adam Mickiewicz University, Tilde and University of Alicante. We focus on efficient implementation of the recurrent deep-learning model as implemented in Amun, the fast inference engine for neural machine translation. We improve the performance with an efficient mini-batching algorithm, and by fusing the softmax operation with the k-best extraction algorithm.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } This paper describes the submissions to the efficiency track for GPUs by members of the University of Edinburgh, Adam Mickiewicz University, Tilde and University of Alicante. We focus on efficient implementation of the recurrent deep-learning model as implemented in Amun, the fast inference engine for neural machine translation. We improve the performance with an efficient mini-batching algorithm, and by fusing the softmax operation with the k-best extraction algorithm. |
Dong, Li; Lapata, Maria Coarse-to-Fine Decoding for Neural Semantic Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. @inproceedings{5887434cbb9143ef9131a1d1641a5d6a, title = {Coarse-to-Fine Decoding for Neural Semantic Parsing}, author = {Li Dong and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {56th Annual Meeting of the Association for Computational Linguistics}, abstract = {Semantic parsing aims at mapping natural language utterances into structured meaning representations. In this work, we propose a structure-aware neural architecture which decomposes the semantic parsing process into two stages. Given an input utterance, we first generate a rough sketch of its meaning, where low-level information (such as variable names and arguments) is glossed over. Then, we fill in missing details by taking into account the natural language input and the sketch itself. Experimental results on four datasets characteristic of different domains and meaning representations show that our approach consistently improves performance, achieving competitive results despite the use of relatively simple decoders.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Semantic parsing aims at mapping natural language utterances into structured meaning representations. In this work, we propose a structure-aware neural architecture which decomposes the semantic parsing process into two stages. Given an input utterance, we first generate a rough sketch of its meaning, where low-level information (such as variable names and arguments) is glossed over. Then, we fill in missing details by taking into account the natural language input and the sketch itself. Experimental results on four datasets characteristic of different domains and meaning representations show that our approach consistently improves performance, achieving competitive results despite the use of relatively simple decoders. |
Bogoychev, Nikolay; Junczys-Dowmunt, Marcin; Heafield, Kenneth; Aji, Alham Accelerating Asynchronous Stochastic Gradient Descent for Neural Machine Translation Inproceedings Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018. @inproceedings{d58b039c6b7f417794c9ffed25ccda90, title = {Accelerating Asynchronous Stochastic Gradient Descent for Neural Machine Translation}, author = {Nikolay Bogoychev and Marcin Junczys-Dowmunt and Kenneth Heafield and Alham Aji}, year = {2018}, date = {2018-01-01}, booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing}, abstract = {In order to extract the best possible performance from asynchronous stochastic gradient descent one must increase the mini-batch size and scale the learning rate accordingly. In order to achieve further speedup we introduce a technique that delays gradient updates effectively increasing the mini-batch size. Unfortunately with the increase of mini-batch size we worsen the stale gradient problem in asynchronous stochastic gradient descent (SGD) which makes the model convergence poor. We introduce local optimizers which mitigate the stale gradient problem and together with fine tuning our momentum we are able to train a shallow machine translation system 27%faster than an optimized baseline with negligible penalty in BLEU.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } In order to extract the best possible performance from asynchronous stochastic gradient descent one must increase the mini-batch size and scale the learning rate accordingly. In order to achieve further speedup we introduce a technique that delays gradient updates effectively increasing the mini-batch size. Unfortunately with the increase of mini-batch size we worsen the stale gradient problem in asynchronous stochastic gradient descent (SGD) which makes the model convergence poor. We introduce local optimizers which mitigate the stale gradient problem and together with fine tuning our momentum we are able to train a shallow machine translation system 27%faster than an optimized baseline with negligible penalty in BLEU. |
Liu, Qianchu; Fancellu, Federico; Webber, Bonnie NegPar: a parallel corpus annotated for negation Inproceedings Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), pp. 3464–3472, European Language Resources Association (ELRA), 2018. @inproceedings{8a39ebf005054173bd9a91e97c7df927, title = {NegPar: a parallel corpus annotated for negation}, author = {Qianchu Liu and Federico Fancellu and Bonnie Webber}, year = {2018}, date = {2018-01-01}, booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)}, pages = {3464--3472}, publisher = {European Language Resources Association (ELRA)}, abstract = {Although the existence of English corpora annotated for negation has allowed for extensive work on monolingual negation detection, little is understood on how negation-related phenomena translate across languages. The current study fills this gap by presenting NegPar, the first English-Chinese parallel corpus annotated for negation in the narrative domain (a collection of stories from Conan Doyle's Sherlock Holmes). While we followed the annotation guidelines in the ConanDoyleNeg corpus (Morante and Daelemans, 2012), we reannotated certain scope-related phenomena to ensure more consistent and interpretable semantic representation. To both ease the annotation process and analyze how similar negation is signaled in the two languages, we experimented with first projecting the annotations from English and then manually correcting the projection output in Chinese. Results show that projecting negation via word-alignment offers limited help to the annotation process, as negation can be rendered in different ways across languages.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Although the existence of English corpora annotated for negation has allowed for extensive work on monolingual negation detection, little is understood on how negation-related phenomena translate across languages. The current study fills this gap by presenting NegPar, the first English-Chinese parallel corpus annotated for negation in the narrative domain (a collection of stories from Conan Doyle's Sherlock Holmes). While we followed the annotation guidelines in the ConanDoyleNeg corpus (Morante and Daelemans, 2012), we reannotated certain scope-related phenomena to ensure more consistent and interpretable semantic representation. To both ease the annotation process and analyze how similar negation is signaled in the two languages, we experimented with first projecting the annotations from English and then manually correcting the projection output in Chinese. Results show that projecting negation via word-alignment offers limited help to the annotation process, as negation can be rendered in different ways across languages. |
Junczys-Dowmunt, Marcin; Heafield, Kenneth; Hoang, Hieu; Grundkiewicz, Roman; Aue, Anthony Marian: Cost-effective High-Quality Neural Machine Translation in C++ Inproceedings The 2nd Workshop on Neural Machine Translation and Generation 2018, 2018. @inproceedings{aef9897a5d21433186e30b8a4b108f9c, title = {Marian: Cost-effective High-Quality Neural Machine Translation in C++}, author = {Marcin Junczys-Dowmunt and Kenneth Heafield and Hieu Hoang and Roman Grundkiewicz and Anthony Aue}, year = {2018}, date = {2018-01-01}, booktitle = {The 2nd Workshop on Neural Machine Translation and Generation 2018}, abstract = {This paper describes the submissions of the “Marian” team to the WNMT 2018 shared task. We investigate combinations of teacher-student training, low-precision matrix products, auto-tuning and other methods to optimize the Transformer model on GPU and CPU. By further integrating these methods with the new averaging attention networks, a recently introduced faster Transformer variant, we create a number of high-quality, high-performance models on the GPU and CPU, dominating the Pareto frontier for this shared task.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } This paper describes the submissions of the “Marian” team to the WNMT 2018 shared task. We investigate combinations of teacher-student training, low-precision matrix products, auto-tuning and other methods to optimize the Transformer model on GPU and CPU. By further integrating these methods with the new averaging attention networks, a recently introduced faster Transformer variant, we create a number of high-quality, high-performance models on the GPU and CPU, dominating the Pareto frontier for this shared task. |
Cheng, Jianpeng; Lapata, Maria Weakly-supervised Neural Semantic Parsing with a Generative Ranker Inproceedings SIGNLL Conference on Computational Natural Language Learning (CoNLL 2018), 2018. @inproceedings{8a5dff2c636e4f289c106463449b89c6, title = {Weakly-supervised Neural Semantic Parsing with a Generative Ranker}, author = {Jianpeng Cheng and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {SIGNLL Conference on Computational Natural Language Learning (CoNLL 2018)}, abstract = {Weakly-supervised semantic parsers are trained on utterance-denotation pairs, treating logical forms as latent. The task is challenging due to the large search space and spuriousness of logical forms. In this paper we introduce a neural parser-ranker system which addresses both challenges based on three innovations: (a) candidate (tree-structured) logical forms in our model are ranked based on two criteria, i.e., whether they are likely to execute to the correct denotation and the degree to which they preserve the meaning of the utterance; (b) a scheduled training procedure effectively balances the contribution of the two objectives; (c) a neurally encoded lexicon is used to inject prior domain knowledge to the model. Experiments on three Freebase datasets demonstrate the effectiveness of our semantic parser, achieving state-of-the-art results.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Weakly-supervised semantic parsers are trained on utterance-denotation pairs, treating logical forms as latent. The task is challenging due to the large search space and spuriousness of logical forms. In this paper we introduce a neural parser-ranker system which addresses both challenges based on three innovations: (a) candidate (tree-structured) logical forms in our model are ranked based on two criteria, i.e., whether they are likely to execute to the correct denotation and the degree to which they preserve the meaning of the utterance; (b) a scheduled training procedure effectively balances the contribution of the two objectives; (c) a neurally encoded lexicon is used to inject prior domain knowledge to the model. Experiments on three Freebase datasets demonstrate the effectiveness of our semantic parser, achieving state-of-the-art results. |
Darwish, Kareem; Mubarak, Hamdy; Eldesouki, Mohamed; AbdelAli, Ahmed; Samih, Younes; Alharbi, Randah; Attia, Mohammed; Magdy, Walid; Kallmeyer, Laura Multi-Dialect Arabic POS Tagging: A CRF Approach Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. @inproceedings{4a5026f61abf4a60b3fe3024330b832db, title = {Multi-Dialect Arabic POS Tagging: A CRF Approach}, author = {Kareem Darwish and Hamdy Mubarak and Mohamed Eldesouki and Ahmed AbdelAli and Younes Samih and Randah Alharbi and Mohammed Attia and Walid Magdy and Laura Kallmeyer}, year = {2018}, date = {2018-01-01}, booktitle = {11th edition of the Language Resources and Evaluation Conference}, abstract = {This paper introduces a new dataset of POS-tagged Arabic tweets in four major dialects along with tagging guidelines. The data, which we are releasing publicly, includes tweets in Egyptian, Levantine, Gulf, and Maghrebi, with 350 tweets for each dialect with appropriate train/test/development splits for 5-fold cross validation. We use a Conditional Random Fields (CRF) sequence labeler to train POS taggers for each dialect and examine the effect of cross and joint dialect training, and give benchmark results for the datasets. Using clitic n-grams, clitic metatypes, and stem templates as features, we were able to train a joint model that can correctly tag four different dialects with an average accuracy of 89.3%.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } This paper introduces a new dataset of POS-tagged Arabic tweets in four major dialects along with tagging guidelines. The data, which we are releasing publicly, includes tweets in Egyptian, Levantine, Gulf, and Maghrebi, with 350 tweets for each dialect with appropriate train/test/development splits for 5-fold cross validation. We use a Conditional Random Fields (CRF) sequence labeler to train POS taggers for each dialect and examine the effect of cross and joint dialect training, and give benchmark results for the datasets. Using clitic n-grams, clitic metatypes, and stem templates as features, we were able to train a joint model that can correctly tag four different dialects with an average accuracy of 89.3%. |
Shao, Yutong; Sennrich, Rico; Bonnie L, ; Fancellu, Federico Evaluating Machine Translation Performance on Chinese Idioms with a Blacklist Method Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. @inproceedings{7fc212d799734d5583bdfeb031a8abafb, title = {Evaluating Machine Translation Performance on Chinese Idioms with a Blacklist Method}, author = {Yutong Shao and Rico Sennrich and {Bonnie L } Webber and Federico Fancellu}, year = {2018}, date = {2018-01-01}, booktitle = {11th edition of the Language Resources and Evaluation Conference}, abstract = {Idiom translation is a challenging problem in machine translation because the meaning of idioms is non-compositional, and a literal(word-by-word) translation is likely to be wrong. In this paper, we focus on evaluating the quality of idiom translation of MT systems. We introduce a new evaluation method based on an idiom-specific blacklist of literal translations, based on the insight that the occurrence of any blacklisted words in the translation output indicates a likely translation error. We introduce a dataset, CIBB (Chinese Idioms Blacklists Bank), and perform an evaluation of a state-of-the-art Chinese→English neural MT system. Our evaluation confirms that a sizable number of idioms in our test set are mistranslated (46.1%), that literal translation error is a common error type, and that our blacklist method is effective at identifying literal translation errors.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Idiom translation is a challenging problem in machine translation because the meaning of idioms is non-compositional, and a literal(word-by-word) translation is likely to be wrong. In this paper, we focus on evaluating the quality of idiom translation of MT systems. We introduce a new evaluation method based on an idiom-specific blacklist of literal translations, based on the insight that the occurrence of any blacklisted words in the translation output indicates a likely translation error. We introduce a dataset, CIBB (Chinese Idioms Blacklists Bank), and perform an evaluation of a state-of-the-art Chinese→English neural MT system. Our evaluation confirms that a sizable number of idioms in our test set are mistranslated (46.1%), that literal translation error is a common error type, and that our blacklist method is effective at identifying literal translation errors. |
Rohde, Hannah; Johnson, Alexander; Schneider, Nathan; Webber, Bonnie 2018, (Cross-Linguistic Discourse Annotation Applications & Perspectives : Final Action Conference ; Conference date: 19-03-2018 Through 21-03-2018). @conference{04d9168e4c1542c580e4df751f831b44, title = {Choosing among alternatives: Conjunction variability comes from both inference and the semantics of discourse adverbials}, author = {Hannah Rohde and Alexander Johnson and Nathan Schneider and Bonnie Webber}, url = {http://textlink.ii.metu.edu.tr/final-action-conference}, year = {2018}, date = {2018-01-01}, pages = {114--121}, note = {Cross-Linguistic Discourse Annotation Applications & Perspectives : Final Action Conference ; Conference date: 19-03-2018 Through 21-03-2018}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Bansal, Sameer; Kamper, Herman; Livescu, Karen; Lopez, Adam; Goldwater, Sharon Low-Resource Speech-to-Text Translation Inproceedings Interspeech 2018, 2018. @inproceedings{0f39a6b3740040e4a24d2ccf23cb873f, title = {Low-Resource Speech-to-Text Translation}, author = {Sameer Bansal and Herman Kamper and Karen Livescu and Adam Lopez and Sharon Goldwater}, year = {2018}, date = {2018-01-01}, booktitle = {Interspeech 2018}, abstract = {Speech-to-text translation has many potential applications for low-resource languages, but the typical approach of cascading speech recognition with machine translation is often impossible, since the transcripts needed to train a speech recognizer are usually not available for low-resource languages. Recent work has found that neural encoder-decoder models can learn to directly translate foreign speech in high-resource scenarios, without the need for intermediate transcription. We investigate whether this approach also works in settings where both data and computation are limited. To make the approach efficient, we make several architectural changes, including a change from character-level to word-level decoding. We find that this choice yields crucial speed improvements that allow us to train with fewer computational resources, yet still performs well on frequent words. We explore models trained on between 20 and 160 hours of data, and find that although models trained on less data have considerably lower BLEU scores, they can still predict words with relatively high precision and recall—around 50% for a model trained on 50 hours of data, versus around 60% for the full 160 hour model. Thus, they may still be useful for some low-resource scenarios.Index Terms: speech translation, low-resource speech processing, encoder-decoder models}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Speech-to-text translation has many potential applications for low-resource languages, but the typical approach of cascading speech recognition with machine translation is often impossible, since the transcripts needed to train a speech recognizer are usually not available for low-resource languages. Recent work has found that neural encoder-decoder models can learn to directly translate foreign speech in high-resource scenarios, without the need for intermediate transcription. We investigate whether this approach also works in settings where both data and computation are limited. To make the approach efficient, we make several architectural changes, including a change from character-level to word-level decoding. We find that this choice yields crucial speed improvements that allow us to train with fewer computational resources, yet still performs well on frequent words. We explore models trained on between 20 and 160 hours of data, and find that although models trained on less data have considerably lower BLEU scores, they can still predict words with relatively high precision and recall—around 50% for a model trained on 50 hours of data, versus around 60% for the full 160 hour model. Thus, they may still be useful for some low-resource scenarios.Index Terms: speech translation, low-resource speech processing, encoder-decoder models |
AbdelRahim A, ; Mubarak, Hamdy; Magdy, Walid An Arabic Speech-Act and Sentiment Corpus of Tweets Inproceedings rd Workshop on Open-Source Arabic Corpora and Processing Tools (part of LREC 2018), 2018. @inproceedings{d3cb00d902eb44a0a0d01b794e862787b, title = {An Arabic Speech-Act and Sentiment Corpus of Tweets}, author = {{AbdelRahim A } Elmadany and Hamdy Mubarak and Walid Magdy}, year = {2018}, date = {2018-01-01}, booktitle = {rd Workshop on Open-Source Arabic Corpora and Processing Tools (part of LREC 2018)}, abstract = {Speech acts are the type of communicative acts within a conversation. Speech act recognition (aka classification) has been an active research in recent years. However, much less attention was directed towards this task in Arabic due to the lack of resources for training an Arabic speech-act classifier. In this paper we present ArSAS , an Arabic corpus of tweets annotated for the tasks of speech-act recognition and sentiment analysis. A large set of 21k Arabic tweets covering multiple topics were collected, prepared and annotated for six different classes of speech-act labels, such as expression, assertion, and question. In addition, the same set of tweets were also annotated with four classes of sentiment. We aim to have this corpus promoting the research in both speech-act recognition and sentiment analysis tasks for Arabic language.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Speech acts are the type of communicative acts within a conversation. Speech act recognition (aka classification) has been an active research in recent years. However, much less attention was directed towards this task in Arabic due to the lack of resources for training an Arabic speech-act classifier. In this paper we present ArSAS , an Arabic corpus of tweets annotated for the tasks of speech-act recognition and sentiment analysis. A large set of 21k Arabic tweets covering multiple topics were collected, prepared and annotated for six different classes of speech-act labels, such as expression, assertion, and question. In addition, the same set of tweets were also annotated with four classes of sentiment. We aim to have this corpus promoting the research in both speech-act recognition and sentiment analysis tasks for Arabic language. |
Guillou, Liane; Hardmeier, Christian Automatic Reference-Based Evaluation of Pronoun Translation Misses the Point Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, Association for Computational Linguistics, 2018. @inproceedings{8bef4d6eed1544f78f93cf924a397792, title = {Automatic Reference-Based Evaluation of Pronoun Translation Misses the Point}, author = {Liane Guillou and Christian Hardmeier}, year = {2018}, date = {2018-01-01}, booktitle = {2018 Conference on Empirical Methods in Natural Language Processing}, publisher = {Association for Computational Linguistics}, abstract = {We compare the performance of the APT and AutoPRF metrics for pronoun translation against a manually annotated dataset comprising human judgements as to the correctness of translations of the PROTEST test suite. Although there is some correlation with the human judgements, a range of issues limit the performance of the automated metrics. Instead, we recommend the use of semiautomatic metrics and test suites in place of fully automatic metrics.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } We compare the performance of the APT and AutoPRF metrics for pronoun translation against a manually annotated dataset comprising human judgements as to the correctness of translations of the PROTEST test suite. Although there is some correlation with the human judgements, a range of issues limit the performance of the automated metrics. Instead, we recommend the use of semiautomatic metrics and test suites in place of fully automatic metrics. |
Mallinson, Jonathan; Sennrich, Rico; Lapata, Maria Sentence Compression for Arbitrary Languages via Multilingual Pivoting Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. @inproceedings{acd94b770ce44de5b93f85a50ca87aca, title = {Sentence Compression for Arbitrary Languages via Multilingual Pivoting}, author = {Jonathan Mallinson and Rico Sennrich and Maria Lapata}, year = {2018}, date = {2018-01-01}, booktitle = {2018 Conference on Empirical Methods in Natural Language Processing}, abstract = {In this paper we advocate the use of bilingual corpora which are abundantly available for training sentence compression models. Our approach borrows much of its machinery from neural machine translation and leverages bilingual pivoting: compressions are obtained by translating a source string into a foreign language and then back-translating it into the source while controlling the translation length. Our model can be trained for any language as long as a bilingual corpus is available and performs arbitrary rewrites without access to compression specific data. We release1 MOSS, a new parallel Multilingual Compression dataset for English, German, and French which can be used to evaluate compression models across languages and genres.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } In this paper we advocate the use of bilingual corpora which are abundantly available for training sentence compression models. Our approach borrows much of its machinery from neural machine translation and leverages bilingual pivoting: compressions are obtained by translating a source string into a foreign language and then back-translating it into the source while controlling the translation length. Our model can be trained for any language as long as a bilingual corpus is available and performs arbitrary rewrites without access to compression specific data. We release1 MOSS, a new parallel Multilingual Compression dataset for English, German, and French which can be used to evaluate compression models across languages and genres. |
Currey, Anna; Heafield, Kenneth Unsupervised Source Hierarchies for Low-Resource Neural Machine Translation Inproceedings NLP Workshop 2018, 2018. @inproceedings{93d73e2e319a4b9c90a919027398b27f, title = {Unsupervised Source Hierarchies for Low-Resource Neural Machine Translation}, author = {Anna Currey and Kenneth Heafield}, year = {2018}, date = {2018-01-01}, booktitle = {NLP Workshop 2018}, abstract = {Incorporating source syntactic information into neural machine translation (NMT) has recently proven successful (Eriguchi et al., 2016; Luong et al., 2016). However, this is generally done using an outside parser to syntactically annotate the training data, making this technique difficult to use for languages or domains for which a reliable parser is not available. In this paper, we introduce an unsupervised tree-to-sequence (tree2seq) model for neural machine translation; this model is able to induce an unsupervised hierarchical structure on the source sentence based on the downstream task of neural machine translation. We adapt the Gumbel tree-LSTM of Choi et al. (2018) to NMT in order to create the encoder.We evaluate our model against sequential and supervised parsing baselines on three low- and medium-resource language pairs. For low-resource cases, the unsupervised tree2seq encoder significantly outperforms the baselines; no improvements are seen for medium-resource translation.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } Incorporating source syntactic information into neural machine translation (NMT) has recently proven successful (Eriguchi et al., 2016; Luong et al., 2016). However, this is generally done using an outside parser to syntactically annotate the training data, making this technique difficult to use for languages or domains for which a reliable parser is not available. In this paper, we introduce an unsupervised tree-to-sequence (tree2seq) model for neural machine translation; this model is able to induce an unsupervised hierarchical structure on the source sentence based on the downstream task of neural machine translation. We adapt the Gumbel tree-LSTM of Choi et al. (2018) to NMT in order to create the encoder.We evaluate our model against sequential and supervised parsing baselines on three low- and medium-resource language pairs. For low-resource cases, the unsupervised tree2seq encoder significantly outperforms the baselines; no improvements are seen for medium-resource translation. |
Gardent, Claire; Narayan, Shashi Deep Learning Approaches to Text Production Conference 2018, (16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT 2018 ; Conference date: 01-06-2018 Through 06-06-2018). @conference{9ef297bcce8a40058d784f2d85eded78, title = {Deep Learning Approaches to Text Production}, author = {Claire Gardent and Shashi Narayan}, url = {http://naacl2018.org/}, doi = {10.18653/v1/N18-6002}, year = {2018}, date = {2018-01-01}, pages = {4--9}, note = {16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT 2018 ; Conference date: 01-06-2018 Through 06-06-2018}, keywords = {}, pubstate = {published}, tppubtype = {conference} } |
Bergmanis, Toms; Goldwater, Sharon Context Sensitive Neural Lemmatization with Lematus Inproceedings 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. @inproceedings{8a31ff7259d345bea99ccd8d4aafc408, title = {Context Sensitive Neural Lemmatization with Lematus}, author = {Toms Bergmanis and Sharon Goldwater}, year = {2018}, date = {2018-01-01}, booktitle = {16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies}, abstract = {The main motivation for developing contextsensitive lemmatizers is to improve performance on unseen and ambiguous words. Yetprevious systems have not carefully evaluated whether the use of context actually helps in these cases. We introduce Lematus, a lemmatizer based on a standard encoder-decoder architecture, which incorporates character-level sentence context. We evaluate its lemmatization accuracy across 20 languages in both a full data setting and a lower-resource setting with 10k training examples in each language. In both settings, we show that including context significantly improves results against a context-free version of the model. Context helps more for ambiguous words than for unseen words, though the latter have a greater effect on overall performance differences between languages. We also compare to three previous context-sensitive lemmatization systems, which all use pre-extracted edit trees as well as hand-selected features and/or additional sources of information such as tagged training data. Without using any of these, our context-sensitive model outperforms the best competitor system (Lemming) in the full-data setting, and performs on par in the lower-resource setting.}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } The main motivation for developing contextsensitive lemmatizers is to improve performance on unseen and ambiguous words. Yetprevious systems have not carefully evaluated whether the use of context actually helps in these cases. We introduce Lematus, a lemmatizer based on a standard encoder-decoder architecture, which incorporates character-level sentence context. We evaluate its lemmatization accuracy across 20 languages in both a full data setting and a lower-resource setting with 10k training examples in each language. In both settings, we show that including context significantly improves results against a context-free version of the model. Context helps more for ambiguous words than for unseen words, though the latter have a greater effect on overall performance differences between languages. We also compare to three previous context-sensitive lemmatization systems, which all use pre-extracted edit trees as well as hand-selected features and/or additional sources of information such as tagged training data. Without using any of these, our context-sensitive model outperforms the best competitor system (Lemming) in the full-data setting, and performs on par in the lower-resource setting. |
2018 |
Improving Machine Translation of Educational Content via Crowdsourcing Inproceedings 11th Edition of the Language Resources and Evaluation Conference, 2018. |
Learning Structured Text Representations Journal Article Transactions of the Association for Computational Linguistics, 6 , pp. 63–76, 2018, ISSN: 2307-387X. |
Bootstrapping Generators from Noisy Data Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |
Whodunnit? Crime Drama as a Case for Natural Language Understanding Journal Article Transactions of the Association for Computational Linguistics, 2018, ISSN: 2307-387X. |
Approaching Neural Grammatical Error Correction as a Low-Resource Machine Translation Task Inproceedings Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pp. 595–606, Association for Computational Linguistics, 2018. |
Discourse Representation Structure Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, pp. 429–439, Association for Computational Linguistics (ACL), 2018. |
Improving Entity Linking by Modeling Latent Relations between Mentions Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Evaluating Discourse Phenomena in Neural Machine Translation Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |
An Evaluation of Image-based Verb Prediction Models against Human Eye-tracking Data Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 758–763, Association for Computational Linguistics, 2018. |
Inducing a lexicon of sociolinguistic variables from code-mixed text Inproceedings 2018 The 4th Workshop on Noisy User-generated Text (W-NUT), 2018. |
Deep Dungeons and Dragons: Learning Character-Action Interactions from Role-Playing Game Transcripts Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |
Confidence Modeling for Neural Semantic Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Neural Machine Translation Techniques for Named Entity Transliteration Inproceedings NEWS 2018: The Seventh Named Entities Workshop, 2018. |
A Structured Syntax-Semantics Interface for English-AMR Alignment Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 1169–1180, Association for Computational Linguistics, 2018. |
What’s This Movie About? A Joint Neural Network Architecture for Movie Content Analysis Inproceedings Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 1770–1781, Association for Computational Linguistics (ACL), 2018. |
Text Mining Brain Imaging Reports Conference 2018, (UK Healthcare Text Analytics Conference (HealTAC-2018) : Unlocking Evidence Contained in Healthcare Free-text, HealTAC-2018 ; Conference date: 18-04-2018 Through 19-04-2018). |
ClassStrength v2: An Adaptive Multilingual Tool for Tweet Classification Inproceedings 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), 2018. |
Abstract Meaning Representation for Paraphrase Detection Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 442–452, Association for Computational Linguistics, 2018. |
Marian: Fast Neural Machine Translation in C++ Inproceedings The 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Structured Alignment Networks for Matching Sentences Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. |
Embedding Words as Distributions with a Bayesian Skip-gram Model Inproceedings 27th International Conference on Computational Linguistics, 2018. |
Canonical Correlation Inference for Mapping Abstract Scenes to Text Inproceedings Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18), 2018. |
Assessing multilingual multimodal image description: Studies of native speaker preferences and translator choices Journal Article Natural Language Engineering, 24 (3), pp. 393–413, 2018, ISSN: 1351-3249, (Route 1). |
BroDyn’18: Workshop on analysis of broad dynamic topics over social media Inproceedings Advances in Information Retrieval - 40th European Conference on IR Research, ECIR 2018, Proceedings, pp. 829–830, Springer-Verlag, 2018, ISBN: 978-3-319-76940-0. |
Document Modeling with External Attention for Sentence Extraction Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Modeling Relational Data with Graph Convolutional Networks Inproceedings The 15th Extended Semantic Web Conference (ESWC), 2018. |
Context-Aware Neural Machine Translation Learns Anaphora Resolution Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Part-of-Speech Tagging for Arabic Gulf Dialect Using Bi-LSTM Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. |
Ranking Sentences for Extractive Summarization with Reinforcement Learning Inproceedings The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |
Multiple Instance Learning Networks for Fine-Grained Sentiment Analysis Journal Article Transactions of the Association for Computational Linguistics, 6 , pp. 17–32, 2018, ISSN: 2307-387X. |
Neural Latent Extractive Document Summarization Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. |
Self-Representation on Twitter Using Emoji Skin Color Modifiers Inproceedings 12th International AAAI Conference on Web and Social Media (ICWSM-18), 2018. |
Evaluating Historical Text Normalization Systems: How Well Do They Generalize? Inproceedings 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |
Summarizing Opinions: Aspect Extraction Meets Sentiment Prediction and They Are Both Weakly Supervised Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. |
Fast Neural Machine Translation Implementation Inproceedings The 2nd Workshop on Neural Machine Translation and Generation 2018, 2018. |
Coarse-to-Fine Decoding for Neural Semantic Parsing Inproceedings 56th Annual Meeting of the Association for Computational Linguistics, 2018. |
Accelerating Asynchronous Stochastic Gradient Descent for Neural Machine Translation Inproceedings Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018. |
NegPar: a parallel corpus annotated for negation Inproceedings Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), pp. 3464–3472, European Language Resources Association (ELRA), 2018. |
Marian: Cost-effective High-Quality Neural Machine Translation in C++ Inproceedings The 2nd Workshop on Neural Machine Translation and Generation 2018, 2018. |
Weakly-supervised Neural Semantic Parsing with a Generative Ranker Inproceedings SIGNLL Conference on Computational Natural Language Learning (CoNLL 2018), 2018. |
Multi-Dialect Arabic POS Tagging: A CRF Approach Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. |
Evaluating Machine Translation Performance on Chinese Idioms with a Blacklist Method Inproceedings 11th edition of the Language Resources and Evaluation Conference, 2018. |
2018, (Cross-Linguistic Discourse Annotation Applications & Perspectives : Final Action Conference ; Conference date: 19-03-2018 Through 21-03-2018). |
Low-Resource Speech-to-Text Translation Inproceedings Interspeech 2018, 2018. |
An Arabic Speech-Act and Sentiment Corpus of Tweets Inproceedings rd Workshop on Open-Source Arabic Corpora and Processing Tools (part of LREC 2018), 2018. |
Automatic Reference-Based Evaluation of Pronoun Translation Misses the Point Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, Association for Computational Linguistics, 2018. |
Sentence Compression for Arbitrary Languages via Multilingual Pivoting Inproceedings 2018 Conference on Empirical Methods in Natural Language Processing, 2018. |
Unsupervised Source Hierarchies for Low-Resource Neural Machine Translation Inproceedings NLP Workshop 2018, 2018. |
Deep Learning Approaches to Text Production Conference 2018, (16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT 2018 ; Conference date: 01-06-2018 Through 06-06-2018). |
Context Sensitive Neural Lemmatization with Lematus Inproceedings 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018. |