Grammatical Error Correction for Basque through a seq2seq neural architecture and synthetic examples

espanolLas arquitecturas neuronales secuencia a secuencia constituyen el estado del arte para abordar la tarea de correccion de errores gramaticales. Sin embargo, su entrenamiento requiere de grandes conjuntos de datos. Este trabajo estudia el uso de modelos neuronales secuencia a secuencia para la correccion de errores gramaticales en euskera. Al no existir datos de entrenamiento para este idioma, hemos desarrollado un metodo basado en reglas para generar de forma sintetica oraciones gramaticalmente incorrectas a partir de una coleccion de oraciones correctas extraidas de un corpus de 500.000 noticias en euskera. Hemos construido diferentes conjuntos de datos de entrenamiento de acuerdo a distintas estrategias para combinar los ejemplos sinteticos. A partir de estos conjuntos de datos hemos entrenado sendos modelos basados en la arquitectura Transformer que hemos evaluado y comparado de acuerdo a las metricas de precision, cobertura y F0.5. Los resultados obtenidos con el mejor modelo alcanzan un F0.5 de 0.87. EnglishSequence-to-sequence neural architectures are the state of the art for addressing the task of correcting grammatical errors. However, large training datasets are required for this task. This paper studies the use of sequence-to-sequence neural models for the correction of grammatical errors in Basque. As there is no training data for this language, we have developed a rule-based method to generate grammatically incorrect sentences from a collection of correct sentences extracted from a corpus of 500,000 news in Basque. We have built different training datasets according to different strategies to combine the synthetic examples. From these datasets different models based on the Transformer architecture have been trained and evaluated according to accuracy, recall and F0.5 score. The results obtained with the best model reach 0.87 of F0.5 score.

[1]  Maite Oronoz Anchordoqui.  Euskarazko errore sintaktikoak detektatzeko eta zuzentzeko baliabideen garapena: datak, postposizio-lokuzioak eta komunztadura , 2009 .

[2]  Matt Post,et al.  Reassessing the Goals of Grammatical Error Correction: Fluency Instead of Grammaticality , 2016, TACL.

[3]  Lukasz Kaiser,et al.  Attention is All you Need , 2017, NIPS.

[4]  Marcin Junczys-Dowmunt,et al.  Approaching Neural Grammatical Error Correction as a Low-Resource Machine Translation Task , 2018, NAACL.

[5]  Noam M. Shazeer,et al.  Corpora Generation for Grammatical Error Correction , 2019, NAACL.

[6]  J. M. Arriola,et al.  Combining Stochastic and Rule-Based Methods for Disambiguation in Agglutinative Languages , 1998, ACL.

[7]  Hitoshi Isahara,et al.  Automatic Error Detection in the Japanese Learners’ English Spoken Data , 2003, ACL.

[8]  Michael Gamon,et al.  Correcting ESL Errors Using Phrasal SMT Techniques , 2006, ACL.

[9]  Marcin Junczys-Dowmunt,et al.  The WikEd Error Corpus: A Corpus of Corrective Wikipedia Edits and Its Application to Grammatical Error Correction , 2014, PolTAL.

[10]  Dan Roth,et al.  Grammatical Error Correction: Machine Translation and Classifiers , 2016, ACL.

[11]  Rico Sennrich,et al.  Edinburgh Neural Machine Translation Systems for WMT 16 , 2016, WMT.

[12]  Zheng Yuan,et al.  Constrained Grammatical Error Correction using Statistical Machine Translation , 2013, CoNLL Shared Task.

[13]  Ming Zhou,et al.  Fluency Boost Learning and Inference for Neural Grammatical Error Correction , 2018, ACL.

[14]  Helen Yannakoudakis,et al.  Auxiliary Objectives for Neural Error Detection Models , 2017, BEA@EMNLP.

[15]  Michael Gamon,et al.  Using Mostly Native Data to Correct Errors in Learners’ Writing , 2010, NAACL.

[16]  H. Ng,et al.  A Multilayer Convolutional Encoder-Decoder Neural Network for Grammatical Error Correction , 2018, AAAI.