Pub Date : 2020-07-01DOI: 10.18653/v1/2020.sigmorphon-1.5
Xiang Yu, Ngoc Thang Vu, Jonas Kuhn
We present an iterative data augmentation framework, which trains and searches for an optimal ensemble and simultaneously annotates new training data in a self-training style. We apply this framework on two SIGMORPHON 2020 shared tasks: grapheme-to-phoneme conversion and morphological inflection. With very simple base models in the ensemble, we rank the first and the fourth in these two tasks. We show in the analysis that our system works especially well on low-resource languages.
{"title":"Ensemble Self-Training for Low-Resource Languages: Grapheme-to-Phoneme Conversion and Morphological Inflection","authors":"Xiang Yu, Ngoc Thang Vu, Jonas Kuhn","doi":"10.18653/v1/2020.sigmorphon-1.5","DOIUrl":"https://doi.org/10.18653/v1/2020.sigmorphon-1.5","url":null,"abstract":"We present an iterative data augmentation framework, which trains and searches for an optimal ensemble and simultaneously annotates new training data in a self-training style. We apply this framework on two SIGMORPHON 2020 shared tasks: grapheme-to-phoneme conversion and morphological inflection. With very simple base models in the ensemble, we rank the first and the fourth in these two tasks. We show in the analysis that our system works especially well on low-resource languages.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126106904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-06-20DOI: 10.18653/v1/2020.sigmorphon-1.1
Ekaterina Vylomova, Jennifer C. White, Elizabeth Salesky, Sabrina J. Mielke, Shijie Wu, E. Ponti, R. Maudslay, Ran Zmigrod, Josef Valvoda, S. Toldova, Francis M. Tyers, E. Klyachko, I. Yegorov, Natalia Krizhanovsky, Paula Czarnowska, Irene Nikkarinen, Andrew Krizhanovsky, Tiago Pimentel, Lucas Torroba Hennigen, Christo Kirov, Garrett Nicolai, Adina Williams, Antonios Anastasopoulos, Hilaria Cruz, Eleanor Chodroff, Ryan Cotterell, Miikka Silfverberg, Mans Hulden
A broad goal in natural language processing (NLP) is to develop a system that has the capacity to process any natural language. Most systems, however, are developed using data from just one language such as English. The SIGMORPHON 2020 shared task on morphological reinflection aims to investigate systems’ ability to generalize across typologically distinct languages, many of which are low resource. Systems were developed using data from 45 languages and just 5 language families, fine-tuned with data from an additional 45 languages and 10 language families (13 in total), and evaluated on all 90 languages. A total of 22 systems (19 neural) from 10 teams were submitted to the task. All four winning systems were neural (two monolingual transformers and two massively multilingual RNN-based models with gated attention). Most teams demonstrate utility of data hallucination and augmentation, ensembles, and multilingual training for low-resource languages. Non-neural learners and manually designed grammars showed competitive and even superior performance on some languages (such as Ingrian, Tajik, Tagalog, Zarma, Lingala), especially with very limited data. Some language families (Afro-Asiatic, Niger-Congo, Turkic) were relatively easy for most systems and achieved over 90% mean accuracy while others were more challenging.
{"title":"SIGMORPHON 2020 Shared Task 0: Typologically Diverse Morphological Inflection","authors":"Ekaterina Vylomova, Jennifer C. White, Elizabeth Salesky, Sabrina J. Mielke, Shijie Wu, E. Ponti, R. Maudslay, Ran Zmigrod, Josef Valvoda, S. Toldova, Francis M. Tyers, E. Klyachko, I. Yegorov, Natalia Krizhanovsky, Paula Czarnowska, Irene Nikkarinen, Andrew Krizhanovsky, Tiago Pimentel, Lucas Torroba Hennigen, Christo Kirov, Garrett Nicolai, Adina Williams, Antonios Anastasopoulos, Hilaria Cruz, Eleanor Chodroff, Ryan Cotterell, Miikka Silfverberg, Mans Hulden","doi":"10.18653/v1/2020.sigmorphon-1.1","DOIUrl":"https://doi.org/10.18653/v1/2020.sigmorphon-1.1","url":null,"abstract":"A broad goal in natural language processing (NLP) is to develop a system that has the capacity to process any natural language. Most systems, however, are developed using data from just one language such as English. The SIGMORPHON 2020 shared task on morphological reinflection aims to investigate systems’ ability to generalize across typologically distinct languages, many of which are low resource. Systems were developed using data from 45 languages and just 5 language families, fine-tuned with data from an additional 45 languages and 10 language families (13 in total), and evaluated on all 90 languages. A total of 22 systems (19 neural) from 10 teams were submitted to the task. All four winning systems were neural (two monolingual transformers and two massively multilingual RNN-based models with gated attention). Most teams demonstrate utility of data hallucination and augmentation, ensembles, and multilingual training for low-resource languages. Non-neural learners and manually designed grammars showed competitive and even superior performance on some languages (such as Ingrian, Tajik, Tagalog, Zarma, Lingala), especially with very limited data. Some language families (Afro-Asiatic, Niger-Congo, Turkic) were relatively easy for most systems and achieved over 90% mean accuracy while others were more challenging.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124253652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-05-27DOI: 10.18653/v1/2020.sigmorphon-1.28
C. Cathcart, Florian Wandl
This paper investigates the ability of neural network architectures to effectively learn diachronic phonological generalizations in amultilingual setting. We employ models using three different types of language embedding (dense, sigmoid, and straight-through). We find that the Straight-Through model out-performs the other two in terms of accuracy, but the Sigmoid model’s language embeddings show the strongest agreement with the traditional subgrouping of the Slavic languages. We find that the Straight-Through model has learned coherent, semi-interpretable information about sound change, and outline directions for future research.
{"title":"In search of isoglosses: continuous and discrete language embeddings in Slavic historical phonology","authors":"C. Cathcart, Florian Wandl","doi":"10.18653/v1/2020.sigmorphon-1.28","DOIUrl":"https://doi.org/10.18653/v1/2020.sigmorphon-1.28","url":null,"abstract":"This paper investigates the ability of neural network architectures to effectively learn diachronic phonological generalizations in amultilingual setting. We employ models using three different types of language embedding (dense, sigmoid, and straight-through). We find that the Straight-Through model out-performs the other two in terms of accuracy, but the Sigmoid model’s language embeddings show the strongest agreement with the traditional subgrouping of the Slavic languages. We find that the Straight-Through model has learned coherent, semi-interpretable information about sound change, and outline directions for future research.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"104 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115690591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-05-25DOI: 10.18653/v1/2020.sigmorphon-1.9
Manuel Mager, Katharina Kann
In this paper, we present the systems of the University of Stuttgart IMS and the University of Colorado Boulder (IMS--CUBoulder) for SIGMORPHON 2020 Task 2 on unsupervised morphological paradigm completion (Kann et al., 2020). The task consists of generating the morphological paradigms of a set of lemmas, given only the lemmas themselves and unlabeled text. Our proposed system is a modified version of the baseline introduced together with the task. In particular, we experiment with substituting the inflection generation component with an LSTM sequence-to-sequence model and an LSTM pointer-generator network. Our pointer-generator system obtains the best score of all seven submitted systems on average over all languages, and outperforms the official baseline, which was best overall, on Bulgarian and Kannada.
{"title":"The IMS–CUBoulder System for the SIGMORPHON 2020 Shared Task on Unsupervised Morphological Paradigm Completion","authors":"Manuel Mager, Katharina Kann","doi":"10.18653/v1/2020.sigmorphon-1.9","DOIUrl":"https://doi.org/10.18653/v1/2020.sigmorphon-1.9","url":null,"abstract":"In this paper, we present the systems of the University of Stuttgart IMS and the University of Colorado Boulder (IMS--CUBoulder) for SIGMORPHON 2020 Task 2 on unsupervised morphological paradigm completion (Kann et al., 2020). The task consists of generating the morphological paradigms of a set of lemmas, given only the lemmas themselves and unlabeled text. Our proposed system is a modified version of the baseline introduced together with the task. In particular, we experiment with substituting the inflection generation component with an LSTM sequence-to-sequence model and an LSTM pointer-generator network. Our pointer-generator system obtains the best score of all seven submitted systems on average over all languages, and outperforms the official baseline, which was best overall, on Bulgarian and Kannada.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130824490","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-27DOI: 10.18653/v1/2020.sigmorphon-1.25
Oliver Adams, Matthew Wiesner, J. Trmal, Garrett Nicolai, David Yarowsky
We investigate the problem of searching for a lexeme-set in speech by searching for its inflectional variants. Experimental results indicate how lexeme-set search performance changes with the number of hypothesized inflections, while ablation experiments highlight the relative importance of different components in the lexeme-set search pipeline and the value of using curated inflectional paradigms. We provide a recipe and evaluation set for the community to use as an extrinsic measure of the performance of inflection generation approaches.
{"title":"Induced Inflection-Set Keyword Search in Speech","authors":"Oliver Adams, Matthew Wiesner, J. Trmal, Garrett Nicolai, David Yarowsky","doi":"10.18653/v1/2020.sigmorphon-1.25","DOIUrl":"https://doi.org/10.18653/v1/2020.sigmorphon-1.25","url":null,"abstract":"We investigate the problem of searching for a lexeme-set in speech by searching for its inflectional variants. Experimental results indicate how lexeme-set search performance changes with the number of hypothesized inflections, while ablation experiments highlight the relative importance of different components in the lexeme-set search pipeline and the value of using curated inflectional paradigms. We provide a recipe and evaluation set for the community to use as an extrinsic measure of the performance of inflection generation approaches.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"102 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133354661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We present a system for morphological reinflection based on an encoder-decoder neural network model with extra convolutional layers. In spite of its simplicity, the method performs reasonably well on all the languages of the SIGMORPHON 2016 shared task, particularly for the most challenging problem of limited-resources reinflection (track 2, task 3). We also find that using only convolution achieves surprisingly good results in this task, surpassing the accuracy of our encoder-decoder model for several languages.
{"title":"Morphological reinflection with convolutional neural networks","authors":"Robert Östling","doi":"10.18653/v1/W16-2003","DOIUrl":"https://doi.org/10.18653/v1/W16-2003","url":null,"abstract":"We present a system for morphological reinflection based on an encoder-decoder neural network model with extra convolutional layers. In spite of its simplicity, the method performs reasonably well on all the languages of the SIGMORPHON 2016 shared task, particularly for the most challenging problem of limited-resources reinflection (track 2, task 3). We also find that using only convolution achieves surprisingly good results in this task, surpassing the accuracy of our encoder-decoder model for several languages.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114534700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many people are multilingual and they may draw from multiple language varieties when writing their messages. This paper is a first step towards analyzing and detecting code-switching within words. We first segment words into smaller units. Then, words are identified that are composed of sequences of subunits associated with different languages. We demonstrate our method on Twitter data in which both Dutch and dialect varieties labeled as Limburgish, a minority language, are used.
{"title":"Automatic Detection of Intra-Word Code-Switching","authors":"Dong Nguyen, L. Cornips","doi":"10.18653/v1/W16-2013","DOIUrl":"https://doi.org/10.18653/v1/W16-2013","url":null,"abstract":"Many people are multilingual and they may draw from multiple language varieties when writing their messages. This paper is a first step towards analyzing and detecting code-switching within words. We first segment words into smaller units. Then, words are identified that are composed of sequences of subunits associated with different languages. We demonstrate our method on Twitter data in which both Dutch and dialect varieties labeled as Limburgish, a minority language, are used.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"52 4","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121016308","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper presents an approach for the formal representation of compo- nents in German compounds. We as- sume that such a formal representa- tion will support the segmentation and analysis of unseen compounds that feature components already seen in other compounds. An extensive lan- guage resource that explicitly codes components of compounds is Ger- maNet, a lexical semantic network for German. We summarize the Ger- maNet approach to the description of compounds, discussing some of its shortcomings. Our proposed exten- sion of this representation builds on the lemon lexicon model for ontolo- gies, established by the W3C Ontol- ogy Lexicon Community Group.
{"title":"Towards a Formal Representation of Components of German Compounds","authors":"Thierry Declerck, P. Lendvai","doi":"10.18653/v1/W16-2017","DOIUrl":"https://doi.org/10.18653/v1/W16-2017","url":null,"abstract":"This paper presents an approach for the formal representation of compo- nents in German compounds. We as- sume that such a formal representa- tion will support the segmentation and analysis of unseen compounds that feature components already seen in other compounds. An extensive lan- guage resource that explicitly codes components of compounds is Ger- maNet, a lexical semantic network for German. We summarize the Ger- maNet approach to the description of compounds, discussing some of its shortcomings. Our proposed exten- sion of this representation builds on the lemon lexicon model for ontolo- gies, established by the W3C Ontol- ogy Lexicon Community Group.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131293531","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper presents MED, the main system of the LMU team for the SIGMORPHON 2016 Shared Task on Morphological Reinflection as well as an extended analysis of how different design choices contribute to the final performance. We model the task of morphological reinflection using neural encoder-decoder models together with an encoding of the input as a single sequence of the morphological tags of the source and target form as well as the sequence of letters of the source form. The Shared Task consists of three subtasks, three different tracks and covers 10 different languages to encourage the use of language-independent approaches. MED was the system with the overall best performance, demonstrating our method generalizes well for the low-resource setting of the SIGMORPHON 2016 Shared Task.
{"title":"MED: The LMU System for the SIGMORPHON 2016 Shared Task on Morphological Reinflection","authors":"Katharina Kann, Hinrich Schütze","doi":"10.18653/v1/W16-2010","DOIUrl":"https://doi.org/10.18653/v1/W16-2010","url":null,"abstract":"This paper presents MED, the main system of the LMU team for the SIGMORPHON 2016 Shared Task on Morphological Reinflection as well as an extended analysis of how different design choices contribute to the final performance. We model the task of morphological reinflection using neural encoder-decoder models together with an encoding of the input as a single sequence of the morphological tags of the source and target form as well as the sequence of letters of the source form. The Shared Task consists of three subtasks, three different tracks and covers 10 different languages to encourage the use of language-independent approaches. MED was the system with the overall best performance, demonstrating our method generalizes well for the low-resource setting of the SIGMORPHON 2016 Shared Task.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126759944","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This work examines CRF-based sequence alignment models for learning natural language morphology. Although these systems have performed well for a limited number of languages, this work, as part of the SIGMORPHON 2016 shared task, specifically sets out to determine whether these models handle non-concatenative morphology as well as previous work might suggest. Results, however, indicate a strong preference for simpler, concatenative morphological systems.
{"title":"Evaluating Sequence Alignment for Learning Inflectional Morphology","authors":"David L. King","doi":"10.18653/v1/W16-2008","DOIUrl":"https://doi.org/10.18653/v1/W16-2008","url":null,"abstract":"This work examines CRF-based sequence alignment models for learning natural language morphology. Although these systems have performed well for a limited number of languages, this work, as part of the SIGMORPHON 2016 shared task, specifically sets out to determine whether these models handle non-concatenative morphology as well as previous work might suggest. Results, however, indicate a strong preference for simpler, concatenative morphological systems.","PeriodicalId":186158,"journal":{"name":"Special Interest Group on Computational Morphology and Phonology Workshop","volume":"352 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122294913","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：