Modeling Target-Side Inflection in Neural Machine Translation

By Ales Tamchyna, Marion Weller-Di Marco, Alexander M. Fraser. 2017

49 citations recorded by the source corpus.

NMT systems have problems with large vocabulary sizes. Byte-pair encoding (BPE) is a popular approach to solving this problem, but while BPE allows the system to generate any target-side word, it does not enable effective generalization over the rich vocabulary in morphologically rich languages with strong inflectional phenomena. We introduce a simple approach to overcome this problem by training a system to produce the lemma of a word and its morphologically rich POS tag, which is then followed by a deterministic generation step. We apply this strategy for English-Czech and English-German translation scenarios, obtaining improvements in both settings. We furthermore show that the improvement is not due to only adding explicit morphological information.

Subfields: Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

1. Filip Jurvc'ivcek
2. Marine Carpuat
3. Weijia Xu
4. Rui Dong
5. Yating Yang
6. Xiao Li
7. Yu Li
8. Ondrej Dusek

Citations in the corpus

• Neural Generation for Czech: Data and Baselines (2019, positive): “This approach is inspired by similar approaches in phrase-based MT (Bojar, 2007; Toutanova et al., 2008; Fraser, 2009) and was developed in parallel to recent similar experiments with two-step neural MT (Nadejde et al., 2017; Tamchyna et al., 2017).”
• A Diverse Data Augmentation Strategy for Low-Resource Neural Machine Translation (2020, methodology): “employ an encoder–decoder to predict a sequence of interleaving morphological tags and lemmas, then use a morphological generator to generate the final results [19].”
• Neural Machine Translation: A Review (2019, background): “Compound or morpheme splitting can mitigate this issue to a certain extent (Hans & Milton, 2016; Tamchyna et al., 2017).”
• Rule-based Morphological Inflection Improves Neural Terminology Translation (2021, background): “Tamchyna et al. (2017) propose to overcome data sparsity caused by inflection by training NMT models to predict the lemma form and morphological tag of each target word.”