Learning How to Simplify From Explicit Labeling of Complex-Simplified Text Pairs

What it is about

End-to-end simplification models learn from parallel sentence pairs and cannot say what they learned. The authors instead find the simplification operations (deletion, replacement, addition, reordering and others) automatically in a parallel corpus, by aligning complex and simple words, and label each source token with its operation. A bidirectional LSTM (two hidden layers of 100 units, 300-dimensional embeddings) predicts one label per input word. The labelling itself reaches micro-averaged F1 0.70 on the positive labels and accuracy 0.92 over all labels. Only deletion and replacement are applied at test time; replacement uses a supervised lexical simplifier. On 1-to-1 alignments the operation-based system reaches the highest SARI and BLEU among those compared (Moses, Nematus, NTS) when the labels are the automatic silver labels, and copies fewer input sentences. A human evaluation of 100 simplifications by four experts rates grammaticality, meaning preservation and simplicity.

Ideas

• I1. A count-based labeller for Mova: for each verse pair from two translations, align the words (tools/translate/ibm.rs) and label each word as kept, replaced, deleted or added. The label counts give a table of candidate author overrides, each with its reason, instead of a list of unexplained differences.
• I2. Use the deleted and added labels as candidate features for function words (open problem 4): a word like that, unto or a comma that one translation drops is a feature to keep or drop, not noise to ignore.
• I3. Report the residual per operation class (add, delete, replace, reorder), as the paper's error table does, so that the translator's per-class residual has a counterpart in normal mode.

For Mova

The paper's sequence labeller is a neural model and is not for Mova; the labelling step, which is alignment plus counting, is. Problems 4 and 8 of the translator, function words and style deltas, are where the labels would be used. First measure how many verse pairs have deletions against replacements, and run the count twice before reading anything into it.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{alvamanchego2017learning,
  title = {{Learning How to Simplify From Explicit Labeling of Complex-Simplified Text Pairs}},
  author = {Alva-Manchego, Fernando and Bingel, Joachim and Paetzold, Gustavo and Scarton, Carolina and Specia, Lucia},
  booktitle = {Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers)},
  year = {2017},
  url = {https://aclanthology.org/I17-1030/}
}

References

29 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Gustavo Henrique Paetzold and Lucia Specia. 2017. Lexical simplification with neural ranking. Proceedings of EACL, pages 34–40. find
2. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. CoRR, abs/1409.0473. ↗
3. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a method for automatic evaluation of machine translation. In Proceedings of ACL, pages 311–318. find our card
4. Regina Barzilay and Noemie Elhadad. 2003. Sentence alignment for monolingual comparable corpora. In Proceedings of EMNLP, pages 25–32. find
5. Rico Sennrich, Orhan Firat, Kyunghyun Cho, Alexandra Birch, Barry Haddow, Julian Hitschler, Marcin Junczys-Dowmunt, Samuel Läubli, Antonio Valerio Miceli Barone, Jozef Mokry, and Maria Nadejde. 2017. Nematus: a toolkit for neural machine translation. In Proceedings of the Software Demonstrations of EACL, pages 65–68, Valencia, Spain. ACL. find
6. Joachim Bingel and Anders Søgaard. 2016. Text Simplification as Tree Labeling. In Proceedings of ACL, pages 337–343. find
7. Stefan Bott and Horacio Saggion. 2011. An unsupervised alignment algorithm for text simplification corpus construction. In Proceedings of MTTG, pages 20–26. find
8. François Chollet. 2015. Keras. https://github.com/fchollet/keras. ↗
9. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Edinburgh neural machine translation systems for wmt 16. In Proceedings of WMT, pages 371–376. find
10. William Coster and David Kauchak. 2011a. Learning to simplify sentences using wikipedia. In Proceedings of the Workshop on Monolingual Text-To-Text Generation, pages 1–9. find
11. Jason R Smith, Chris Quirk, and Kristina Toutanova. 2010. Extracting parallel sentences from comparable corpora using document level alignment. In Proceedings of NAACL, pages 403–411. find
12. William Coster and David Kauchak. 2011b. Simple English Wikipedia: a new text simplification task. In Proceedings of ACL, pages 665–669. find
13. Matthew Snover, Bonnie Dorr, Richard Schwartz, Linnea Micciulla, and John Makhoul. 2006. A study of translation edit rate with targeted human annotation. In Proceedings of AMTA, pages 223–231. find our card
14. John Duchi, Elad Hazan, and Yoram Singer. 2011. Adaptive subgradient methods for online learning and stochastic optimization. Journal of Machine Learning Research, 12:2121–2159. find
15. Lucia Specia. 2010. Translating from complex to simplified sentences. In Computational Processing of the Portuguese Language, pages 30–39. Springer. find
16. Dan Feblowitz and David Kauchak. 2013. Sentence simplification as tree transduction. In Proceedings of the PITR Workshop, pages 1–10. find
17. Md Sultan, Steven Bethard, and Tamara Sumner. 2014. Back to basics for monolingual alignment: Exploiting word similarity and contextual evidence. TACL, 2:219–230. find
18. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondřej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proceedings of ACL - Demonstration Session, pages 177–180. find
19. Kristian Woodsend and Mirella Lapata. 2011. Learning to simplify sentences with quasi-synchronous grammar and integer programming. In Proceedings of EMNLP, pages 409–420. find our card
20. Sander Wubben, Antal van den Bosch, and Emiel Krahmer. 2012. Sentence simplification by monolingual machine translation. In Proceedings of ACL, pages 1015–1024. find our card
21. Shashi Narayan and Claire Gardent. 2014. Hybrid simplification using deep semantics and machine translation. In Proceedings of ACL, pages 435–445. find
22. Wei Xu, Chris Callison-Burch, and Courtney Napoles. 2015. Problems in current text simplification research: New data can help. TACL, 3:283–297. find
23. Sergiu Nisioi, Sanja Štajner, Simone Paolo Ponzetto, and Liviu P. Dinu. 2017. Exploring neural text simplification models. In Proceedings of ACL, pages 85–91. find
24. Gustavo H. Paetzold and Lucia Specia. 2016a. Semeval 2016 task 11: Complex word identification. In Proceedings of the 10th SemEval, pages 560–569. find
25. Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch. 2016. Optimizing statistical machine translation for text simplification. TACL, 4:401–415. find our card
26. Gustavo Henrique Paetzold and Lucia Specia. 2016b. Vicinity-driven paragraph and sentence alignment for comparable corpora. CoRR, abs/1612.04113. ↗
27. Xingxing Zhang and Mirella Lapata. 2017. Sentence simplification with deep reinforcement learning. In Proceedings of EMNLP, pages 595–605. find
28. Yaoyuan Zhang, Zhenxu Ye, Yansong Feng, Dongyan Zhao, and Rui Yan. 2017. A constrained sequence-to-sequence neural model for sentence simplification. CoRR, abs/1704.02312. ↗
29. Zhemin Zhu, Delphine Bernhard, and Iryna Gurevych. 2010. A monolingual tree-based translation model for sentence simplification. In Proceedings of COL-ING, pages 1353–1361. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/manchego-2017-learning-how-simplify-explicit-labeling/