Controlling the Reading Level of Machine Translation Output

Controlling the Reading Level of Machine Translation Output Kelly Marchisio∗and Jialiang Guo∗ and Cheng-I Lai and Philipp Koehn Center for Language and Speech Processing Johns Hopkins University {kellym,guo,clai24,phi}@jhu.edu Abstract Today’s machine translation systems out- put the same translation for a given in- put, despite important differences between users. In practice, translations should be customized for each reader, for instance when translating for children versus in a business setting. In this paper, we in- troduce the task of reading level control to machine translation, and provide the first results. Our methods can be used to raise or lower the reading level of output translations. In our first approach, source- side sentences in the training corpus are tagged based on the reading level (read- ability) of the matching target sentences. Our second approach alters the traditional encoder-decoder architecture by specify- ing a joint encoder and separate decoders for simple and complex decoding modes, with training data partitioned by reading level. We demonstrate control over output readability score on three test sets in the Spanish–English language direction. 1 Introduction Though the goal of machine translation is to generate semantically accurate translations from one language to another, there are other fac- tors which affect whether a translation is “good”. One often-neglected factor is the reading level of the translation—different contexts require differ- ent reading levels. When translating for less- skilled readers, one may desire a translation with ∗ Equal contribution ∗c⃝ 2019 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. common vocabulary and simple sentence struc- tures. In a professional setting, however, one often requires concise language with advanced vocabu- lary and syntactic structure. For instance, when translating a Spanish web page about machine translation to an English- speaking 7-year-old, one might output, “machine translation is a way to take a sentence from one language and turn it into a sentence in another lan- guage”. When advertising new machine transla- tion software to a potential investor, one might ex- plain, “machine translation is the automated pro- cess by which a sentence in a source language can be converted into a sentence in another lan- guage”. Both sentences carry the same meaning and do not require specialist technical knowledge, but decreasing the complexity in the first makes it easier for a child to understand, and increasing the complexity in the second makes it sound more pro- fessional and sophisticated. Furthermore, for na- tive speakers of low-resource languages where ma- chine translation quality may currently be poor but who can read basic phrases in a second language where translation quality is high, they may prefer to read a lower complexity but semantically accu- rate translation in their second language over an inaccurate, garbled message in their native tongue. In this paper, we introduce the task of reading level control (readability control) to machine trans- lation. We develop two methodologies that con- trol the reading level of a translation in the Span- ish–English language direction, focusing on …

Content truncated to 500 words