Filling Gender & Number Gaps in Neural Machine Translation with Black-box Context Injection

By Amit Moryossef, Roee Aharoni, Yoav Goldberg. 2019

55 citations recorded by the source corpus.

When translating from a language that does not morphologically mark information such as gender and number into a language that does, translation systems must “guess” this missing information, often leading to incorrect translations in the given context. We propose a black-box approach for injecting the missing information to a pre-trained neural machine translation system, allowing to control the morphological variations in the generated translations without changing the underlying model or training data. We evaluate our method on an English to Hebrew translation task, and show that it is effective in injecting the gender and number information and that supplying the correct information improves the translation accuracy in up to 2.3 BLEU on a female-speaker test set for a state-of-the-art online black-box system. Finally, we perform a fine-grained syntactic analysis of the generated translations that shows the effectiveness of our method.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Natural Language Processing

Citations in the corpus

CTRL: A Conditional Transformer Language Model for Controllable Generation (2019, background): “All other control codes are injected into the data without such special treatment (Moryossef et al., 2019; Caswell et al., 2019).”
Societal Biases in Language Generation: Progress and Challenges (2021, methodology): “While Google Translate 3 has been the most popular commercial system to analyze for gender biases (Prates et al., 2019; Moryossef et al., 2019; Stanovsky et al., 2019; Cho et al., 2019; Farkas and N´emeth, 2020), Stanovsky et al. (2019) 2021) also examine Yandex.”
Scalable Cross Lingual Pivots to Model Pronoun Gender for Translation (2020, background): “ quality, while Miculicich Werlen and Popescu-Belis (2017a) presents a proof-of-concept where coreference links are available at decoding time. Most similar to our current work, Habash et al. (2019); Moryossef et al. (2019); Vanmassenhove et al. (2018) improve gender ﬁdelity in translations in of ﬁr”
Reducing Gender Bias in Neural Machine Translation as a Domain Adaptation Problem (2020, neutral): “Moryossef et al. (2019) similarly prepend a short phrase at inference time which acts as a gender domain label for the entire sentence.”
Investigating Failures of Automatic Translation in the Case of Unambiguous Gender (2021, background): “Several works aim to enrich the gender input to an MT system by adding additional gold annotation or context (Stafanovičs et al., 2020; Saunders et al., 2020; Moryossef et al., 2019).”
Gender-Aware Reinflection using Linguistically Enhanced Neural Models (2020, background): “…(Rudinger et al., 2018; Zhao et al., 2018a), machine translation (Rabinovich et al., 2017; Vanmassenhove et al., 2018; Font and Costa-jussà, 2019; Moryossef et al., 2019; Stanovsky et al., 2019; Stafanoviˇcs et al., 2020; Gonen and Webster, 2020), named entity recognition (Mehrabi et al., 2019),…”
Neural Machine Translation Doesn’t Translate Gender Coreference Right Unless You Make It (2020, background): “Moryossef et al. (2019) also incorporate a single explicit gender feature for each sentence at inference.”
First the Worst: Finding Better Gender Translations During Beam Search (2021, background): “…et al., 2019).154 They could also be user-defined for known enti-155 ties (Vanmassenhove et al., 2018), or if translating156 the same sentence with different entity genders to157 produce multiple outputs (Moryossef et al., 2019).158 The inferred case determines entities automati-159 cally.”
Improving Gender Translation Accuracy with Filtered Self-Training (2021, unknown): “…papers address the topic of gender in the context of ambiguous input or propose methods to control for gender or to augment data with gender information (Elaraby et al., 2018; Moryossef et al., 2019; Prates et al., 2018; Saunders et al., 2020; Stafanoviˇcs et al., 2020; Vanmassenhove et al., 2018).”
Mitigating Gender Bias in Machine Translation with Target Gender Annotations (2020, unknown): “Moryossef et al. (2019) undertakes a similar approach at the inference time using phrases (e.g. “she said:” ) that imply the speaker’s gender.”
A Survey on Gender Bias in Natural Language Processing (2021, result): “However, Bentivogli et al. [2020] point out that previously obtained BLEU gains [Moryossef et al. 2019; Vanmassenhove et al. 2018] cannot be ascribed with certainty to a better control of gender features and following previous research [Elaraby et al. 2018; Vanmassenhove et al. 2018] underlie the…”
Gender Bias in Machine Translation (2021, background): “Moryossef et al. (2019) attempt to control the production of feminine references to the speaker and numeral inflections (plural or singular) for the listener(s) in an English-Hebrew spoken language setting.”
Automatically Identifying Gender Issues in Machine Translation using Perturbations (2020, background): “A separate but related line of work focuses on generating correct inflections when translating to gender-marking languages (Vanmassenhove et al., 2018; Moryossef et al., 2019).”