Machine Translation for Machines: the Sentiment Classification Use Case

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1368–1374, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1368 Machine Translation for Machines: the Sentiment Classification Use Case Amirhossein Tebbifakhr1,2, Luisa Bentivogli1, Matteo Negri1, Marco Turchi1 1 Fondazione Bruno Kessler, Via Sommarive 18, Povo, Trento - Italy 2 University of Trento, Italy {atebbifakhr,bentivo,negri,turchi}@fbk.eu Abstract We propose a neural machine translation (NMT) approach that, instead of pursuing ad- equacy and fluency (“human-oriented” qual- ity criteria), aims to generate translations that are best suited as input to a natural language processing component designed for a specific downstream task (a “machine-oriented” crite- rion). Towards this objective, we present a reinforcement learning technique based on a new candidate sampling strategy, which ex- ploits the results obtained on the downstream task as weak feedback. Experiments in senti- ment classification of Twitter data in German and Italian show that feeding an English clas- sifier with machine-oriented translations sig- nificantly improves its performance. Clas- sification results outperform those obtained with translations produced by general-purpose NMT models as well as by an approach based on reinforcement learning. Moreover, our re- sults on both languages approximate the clas- sification accuracy computed on gold standard English tweets. 1 Introduction Traditionally, machine translation (MT) pursues a “human-oriented” objective: generating fluent and adequate output to be consumed by speakers of the target language. But what if the intended use of MT is to feed a natural language processing (NLP) component instead of a human? This, for instance, happens when MT is used as a pre-processing step to perform a downstream NLP task in a language for which dedicated tools are not available due to the scarcity of task-specific training data. The rapid growth of cloud-based software-as-a-service offerings provides a typical example of this sit- uation: a variety of affordable high-performance NLP tools can be easily accessed via APIs but of- ten they are available only for a few languages. Translating into one of these high-resource lan- guages gives the possibility to address the down- stream task by: i) using existing tools for that lan- guage to process the translated text, and ii) pro- jecting their output back to the original language. However, using MT “as is” might not be opti- mal for different reasons. First, despite the qual- itative leap brought by neural networks, MT is still not perfect (Koehn and Knowles, 2017). Sec- ond, previous literature shows that MT can alter some of the properties of the source text (Mirkin et al., 2015; Rabinovich et al., 2017; Vanmassen- hove et al., 2018). Finally, even in the case of a perfect MT able to preserve all the traits of the source sentence, models are still trained on par- allel data, which are created by humans and thus reflect quality criteria relevant for humans. In this work, we posit that these criteria might not be the optimal ones for a machine (i.e. a down- stream NLP …

Content truncated to 500 words