# Explaining Translationese: why are Neural Classifiers Better and what do they Learn?

arXiv: https://arxiv.org/abs/2210.13391 · HTML: https://mlsanity.com/paper/2210.13391/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 46 references that are in the map.
- First posted on arXiv 24 Oct 2022.

## Builds on (most cited first, 18 of 18)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps](https://mlsanity.com/paper/1312.6034/) (2013)
- [Learning Important Features Through Propagating Activation Differences](https://mlsanity.com/paper/1704.02685/) (2017)
- [Captum: A unified and generic model interpretability library for PyTorch](https://mlsanity.com/paper/2009.07896/) (2020)
- [TyDi QA: A Benchmark for Information-Seeking Question Answering in Typologically Diverse Languages](https://mlsanity.com/paper/2003.05002/) (2020)
- [Probing Neural Network Comprehension of Natural Language Arguments](https://mlsanity.com/paper/1907.07355/) (2019)
- [Translation Artifacts in Cross-lingual Transfer Learning](https://mlsanity.com/paper/2004.04721/) (2020)
- [Did the Model Understand the Question?](https://mlsanity.com/paper/1805.05492/) (2018)
- [Attaining the Unattainable? Reassessing Claims of Human Parity in Neural Machine Translation](https://mlsanity.com/paper/1808.10432/) (2018)
- [The Effect of Translationese in Machine Translation Test Sets](https://mlsanity.com/paper/1906.08069/) (2019)
- [Translationese as a Language in "Multilingual" NMT](https://mlsanity.com/paper/1911.03823/) (2019)
- [XLDA: Cross-Lingual Data Augmentation for Natural Language Inference and Question Answering](https://mlsanity.com/paper/1905.11471/) (2019)
- [Personality Trait Detection Using Bagged SVM over BERT Word Embedding Ensembles](https://mlsanity.com/paper/2010.01309/) (2020)
- [Comparing Feature-Engineering and Feature-Learning Approaches for Multilingual Translationese Classification](https://mlsanity.com/paper/2109.07604/) (2021)
- [Unsupervised Identification of Translationese](https://mlsanity.com/paper/1609.03205/) (2016)
- [Incorporating Count-Based Features into Pre-Trained Models for Improved Stance Detection](https://mlsanity.com/paper/2010.09078/) (2020)
- [UoB at SemEval-2020 Task 12: Boosting BERT with Corpus Level Information](https://mlsanity.com/paper/2008.08547/) (2020)

## Built on by (most cited first, 5 of 5)

- [Measuring Spurious Correlation in Classification: 'Clever Hans' in Translationese](https://mlsanity.com/paper/2308.13170/) (2023)
- [Translating away Translationese without Parallel Data](https://mlsanity.com/paper/2310.18830/) (2023)
- [Topic Aware Probing: From Sentence Length Prediction to Idiom Identification how reliant are Neural Language Models on Topic?](https://mlsanity.com/paper/2403.02009/) (2024)
- [Mitigating Translationese in Low-resource Languages: The Storyboard Approach](https://mlsanity.com/paper/2407.10152/) (2024)
- [Translationese-index: Using Likelihood Ratios for Graded and Generalizable Measurement of Translationese](https://mlsanity.com/paper/2507.12260/) (2025)

## Abstract (arXiv)

Recent work has shown that neural feature- and representation-learning, e.g. BERT, achieves superior performance over traditional manual feature engineering based approaches, with e.g. SVMs, in translationese classification tasks. Previous research did not show (i) whether the difference is because of the features, the classifiers or both, and (ii) what the neural classifiers actually learn. To address (i), we carefully design experiments that swap features between BERT- and SVM-based classifiers. We show that an SVM fed with BERT representations performs at the level of the best BERT classifiers, while BERT learning and using handcrafted features performs at the level of an SVM using handcrafted features. This shows that the performance differences are due to the features. To address (ii) we use integrated gradients and find that (a) there is indication that information captured by hand-crafted features is only a subset of what BERT learns, and (b) part of BERT's top performance results are due to BERT learning topic differences and spurious correlations with translationese.