# The paradox of the compositionality of natural language: a neural machine translation case study

arXiv: https://arxiv.org/abs/2108.05885 · HTML: https://mlsanity.com/paper/2108.05885/

## Key facts

- Cited by 59 papers in the map (data through 14 Oct 2026).
- Builds on 28 of its 55 references that are in the map.
- First posted on arXiv 12 Aug 2021.

## Builds on (most cited first, 28 of 28)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [A Call for Clarity in Reporting BLEU Scores](https://mlsanity.com/paper/1804.08771/) (2018)
- [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://mlsanity.com/paper/1904.01038/) (2019)
- [Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference](https://mlsanity.com/paper/1902.01007/) (2019)
- [The FLORES-101 Evaluation Benchmark for Low-Resource and Multilingual Machine Translation](https://mlsanity.com/paper/2106.03193/) (2021)
- [Measuring Compositional Generalization: A Comprehensive Method on Realistic Data](https://mlsanity.com/paper/1912.09713/) (2019)
- [Compositionality decomposed: how do neural networks generalise?](https://mlsanity.com/paper/1908.08351/) (2019)
- [COGS: A Compositional Generalization Challenge Based on Semantic Interpretation](https://mlsanity.com/paper/2010.05465/) (2020)
- [Scaling Neural Machine Translation](https://mlsanity.com/paper/1806.00187/) (2018)
- [Good-Enough Compositional Data Augmentation](https://mlsanity.com/paper/1904.09545/) (2019)
- [Compositional Generalization and Natural Language Variation: Can a Semantic Parsing Approach Handle Both?](https://mlsanity.com/paper/2010.12725/) (2020)
- [Improving Text-to-SQL Evaluation Methodology](https://mlsanity.com/paper/1806.09029/) (2018)
- [The Tatoeba Translation Challenge -- Realistic Data Sets for Low Resource and Multilingual MT](https://mlsanity.com/paper/2010.06354/) (2020)
- [Compositional generalization through meta sequence-to-sequence learning](https://mlsanity.com/paper/1906.05381/) (2019)
- [The emergence of number and syntax units in LSTM language models](https://mlsanity.com/paper/1903.07435/) (2019)
- [Cognitive Science in the era of Artificial Intelligence: A roadmap for reverse-engineering the infant language-learner](https://mlsanity.com/paper/1607.08723/) (2016)
- [Human few-shot learning of compositional instructions](https://mlsanity.com/paper/1901.04587/) (2019)
- [On Compositional Generalization of Neural Machine Translation](https://mlsanity.com/paper/2105.14802/) (2021)
- [Jump to better conclusions: SCAN both left and right](https://mlsanity.com/paper/1809.04640/) (2018)
- [Analysing Neural Language Models: Contextual Decomposition Reveals Default Reasoning in Number and Gender Assignment](https://mlsanity.com/paper/1909.08975/) (2019)
- [Transcoding compositionally: using attention to find more generalizable solutions](https://mlsanity.com/paper/1906.01234/) (2019)
- [On Compositionality in Neural Machine Translation](https://mlsanity.com/paper/1911.01497/) (2019)
- [Can Transformers Jump Around Right in Natural Language? Assessing Performance Transfer from SCAN](https://mlsanity.com/paper/2107.01366/) (2021)
- [LSTMs Compose (and Learn) Bottom-Up](https://mlsanity.com/paper/2010.04650/) (2020)
- [The Unreasonable Volatility of Neural Machine Translation Models](https://mlsanity.com/paper/2005.12398/) (2020)
- [Influence Paths for Characterizing Subject-Verb Number Agreement in LSTM Language Models](https://mlsanity.com/paper/2005.01190/) (2020)
- [Siamese recurrent networks learn first-order logic reasoning and exhibit zero-shot compositional generalization](https://mlsanity.com/paper/1906.00180/) (2019)
- [Sometimes We Want Translationese](https://mlsanity.com/paper/2104.07623/) (2021)

## Built on by (most cited first, 30 of 59)

- [Sudden Drops in the Loss: Syntax Acquisition, Phase Transitions, and Simplicity Bias in MLMs](https://mlsanity.com/paper/2309.07311/) (2023)
- [State-of-the-art generalisation research in NLP: A taxonomy and review](https://mlsanity.com/paper/2210.03050/) (2022)
- [Multi-lingual and Multi-cultural Figurative Language Understanding](https://mlsanity.com/paper/2305.16171/) (2023)
- [Break It Down: Evidence for Structural Compositionality in Neural Networks](https://mlsanity.com/paper/2301.10884/) (2023)
- [LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints](https://mlsanity.com/paper/2410.06458/) (2024)
- [Challenges in Generalization in Open Domain Question Answering](https://mlsanity.com/paper/2109.01156/) (2021)
- [Do GPTs Produce Less Literal Translations?](https://mlsanity.com/paper/2305.16806/) (2023)
- [How BPE Affects Memorization in Transformers](https://mlsanity.com/paper/2110.02782/) (2021)
- [Can Transformer be Too Compositional? Analysing Idiom Processing in Neural Machine Translation](https://mlsanity.com/paper/2205.15301/) (2022)
- [Translate Meanings, Not Just Words: IdiomKB's Role in Optimizing Idiomatic Translation with Language Models](https://mlsanity.com/paper/2308.13961/) (2023)
- [COMPOSER: Compositional Reasoning of Group Activity in Videos with Keypoint-Only Modality](https://mlsanity.com/paper/2112.05892/) (2021)
- [Linking Emergent and Natural Languages via Corpus Transfer](https://mlsanity.com/paper/2203.13344/) (2022)
- [A Survey on Compositional Learning of AI Models: Theoretical and Experimental Practices](https://mlsanity.com/paper/2406.08787/) (2024)
- [On The Specialization of Neural Modules](https://mlsanity.com/paper/2409.14981/) (2024)
- [Benchmarking Compositionality with Formal Languages](https://mlsanity.com/paper/2208.08195/) (2022)
- [Finding Memo: Extractive Memorization in Constrained Sequence Generation Tasks](https://mlsanity.com/paper/2210.12929/) (2022)
- [ANALOGICAL -- A Novel Benchmark for Long Text Analogy Evaluation in Large Language Models](https://mlsanity.com/paper/2305.05050/) (2023)
- [The Validity of Evaluation Results: Assessing Concurrence Across Compositionality Benchmarks](https://mlsanity.com/paper/2310.17514/) (2023)
- [Crossing the Threshold: Idiomatic Machine Translation through Retrieval Augmentation and Loss Weighting](https://mlsanity.com/paper/2310.07081/) (2023)
- [How sensitive are translation systems to extra contexts? Mitigating gender bias in Neural Machine Translation models through relevant contexts](https://mlsanity.com/paper/2205.10762/) (2022)
- [Transparency at the Source: Evaluating and Interpreting Language Models With Access to the True Distribution](https://mlsanity.com/paper/2310.14840/) (2023)
- [Are Representations Built from the Ground Up? An Empirical Examination of Local Composition in Language Models](https://mlsanity.com/paper/2210.03575/) (2022)
- [Automatic Evaluation and Analysis of Idioms in Neural Machine Translation](https://mlsanity.com/paper/2210.04545/) (2022)
- [Explicit Syntactic Guidance for Neural Text Generation](https://mlsanity.com/paper/2306.11485/) (2023)
- [From Frege to chatGPT: Compositionality in language, cognition, and deep neural networks](https://mlsanity.com/paper/2405.15164/) (2024)
- [Learning Disentangled Semantic Spaces of Explanations via Invertible Neural Networks](https://mlsanity.com/paper/2305.01713/) (2023)
- [Vocabulary for Universal Approximation: A Linguistic Perspective of Mapping Compositions](https://mlsanity.com/paper/2305.12205/) (2023)
- [Analyzing the Inner Workings of Transformers in Compositional Generalization](https://mlsanity.com/paper/2502.15277/) (2025)
- [Systematicity, Compositionality and Transitivity of Deep NLP Models: a Metamorphic Testing Perspective](https://mlsanity.com/paper/2204.12316/) (2022)
- [Target-Side Augmentation for Document-Level Machine Translation](https://mlsanity.com/paper/2305.04505/) (2023)

## Abstract (arXiv)

Obtaining human-like performance in NLP is often argued to require compositional generalisation. Whether neural networks exhibit this ability is usually studied by training models on highly compositional synthetic data. However, compositionality in natural language is much more complex than the rigid, arithmetic-like version such data adheres to, and artificial compositionality tests thus do not allow us to determine how neural models deal with more realistic forms of compositionality. In this work, we re-instantiate three compositionality tests from the literature and reformulate them for neural machine translation (NMT). Our results highlight that: i) unfavourably, models trained on more data are more compositional; ii) models are sometimes less compositional than expected, but sometimes more, exemplifying that different levels of compositionality are required, and models are not always able to modulate between them correctly; iii) some of the non-compositional behaviours are mistakes, whereas others reflect the natural variation in data. Apart from an empirical study, our work is a call to action: we should rethink the evaluation of compositionality in neural networks and develop benchmarks using real data to evaluate compositionality on natural language, where composing meaning is not as straightforward as doing the math.