Several papers accepted at EMNLP 2018

Empirical Methods in Natural Language Processing (EMNLP) is a leading conference in the area of Natural Language Processing. This year the NLP group at DIKU has a total of 16 papers accepted at the conference (7 at EMNLP, 4 at CoNLL, 4 at BlackBoxNLP and 1 at WASSA). The research themes across the accepted papers are centred around multitask learning, multilingual learning, and question answering.

For example, PhD Yova Kementchedjhieva and Professor Anders Søgaard together with Sebastian Ruder, PhD at National University of Ireland, and Ryan Cotterell, PhD at Johns Hopkins University, submitted a paper on Generalizing Procrustes Analysis for Better Bilingual Dictionary Induction, which proposes a different take on machine translation. Currently machine translation relies on the availability of parallel data in different languages - think the Rosetta Stone, where the knowledge on a well-known language is used to decode an unknown language. For many smaller languages, this kind of data is scarce or non-existent, which motivates work towards machine translation based on non-parallel monolingual data. The paper introduces an advancement in this field, whereby using Generalized Procrustes Analysis we can achieve state-of-the-art results on the task of Bilingual Dictionary Induction, core to machine translation from non-parallel data.

For more information about the NLP Group’s activities, go to the NLP website.

Full list of papers by DIKU researchers that have been accepted

EMNLP 2018 (Conference on Empirical Methods in Natural Language Processing) 

• Mareike Hartmann, Yova Kementchedjhieva and Anders Søgaard. 
  Why is unsupervised alignment of English embeddings from different algorithms so hard?
  
  
• Ana Valeria Gonzalez-Garduño, Isabelle Augenstein and Anders Søgaard. 
  A strong baseline for question relevancy ranking
  
  
• Sebastian Ruder, Ryan Cotterell, Yova Kementchedjhieva and Anders Søgaard. 
  A Discriminative Latent-Variable Model for Bilingual Lexicon Induction
  
  
• Miryam de Lhoneux, Johannes Bjerva, Isabelle Augenstein and Anders Søgaard. 
  Parameter sharing between dependency parsers for related languages
  
  
• Desmond Elliott. 
  Adversarial Evaluation of Multimodal Translation
  
  
• Mostafa Abdou, Vinit Ravishankar, Artur Kulmizev, Lasha Abzianidze, and Johan Bos. 
  What Can We Learn From Semantic Tagging?
  
  
• Rahul Aralikatte, Neelamadhav Gantayat, Naveen Panwar, Anush Sankaran and Senthil Mani.
  Sanskrit Sandhi Splitting using seq2(seq)^2

CoNLL 2018 (Conference on Computational Natural Language Learning)

• Yova Kementchedjhieva, Sebastian Ruder, Ryan Cotterell and Anders Søgaard. Generalizing Procrustes Analysis for Better Bilingual Dictionary Induction
  
  
• Maria Barrett, Joachim Bingel, Nora Hollenstein, Marek Rei and Anders Søgaard. Sequence classification with human attention. 
  
  
• Á. Kádár, Desmond Elliott, M-A. Côté, G. Chrupała and A. Alishahi. Lessons learned in multilingual grounded language learning.
  
  
• Yova Kementchedjhieva, Johannes Bjerva, Isabelle Augenstein. Copenhagen at CoNLL-SIGMORPHON 2018: Multilingual Inflection in Context with Explicit Morphosyntactic Decoding.

BlackBoxNLP 2018

• Anders Søgaard, Miryam de Lhoneux and Isabelle Augenstein. Nightmare at test time: How punctuation prevents parsers from generalizing. 
  
  
• Emma Kerinec, Chloé Braud and Anders Søgaard. When does deep multi-task learning work for loosely related document classification tasks?
  
  
• Ola Rønning, Daniel Hardt and Anders Søgaard. Ellipsis Resolution in Neural Networks
  
  
• Yova Kementchedjhieva, Adam Lopez. 'Indicatements' that character language models learn English morpho-syntactic units and regularities

WASSA (Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis)

• Jan Lukeš and Anders Søgaard. Sentiment analysis under temporal shift.