# When Informal Text Breaks NLI: Tokenization Failure, Distribution Shift, and Targeted Mitigations

arXiv: https://arxiv.org/abs/2604.16787 · HTML: https://mlsanity.com/paper/2604.16787/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 8 of its 16 references that are in the map.
- First posted on arXiv 18 Apr 2026.

## Builds on (most cited first, 8 of 8)

- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [A large annotated corpus for learning natural language inference](https://mlsanity.com/paper/1508.05326/) (2015)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference](https://mlsanity.com/paper/1902.01007/) (2019)
- [Annotation Artifacts in Natural Language Inference Data](https://mlsanity.com/paper/1803.02324/) (2018)
- [Hypothesis Only Baselines in Natural Language Inference](https://mlsanity.com/paper/1805.01042/) (2018)
- [Competency Problems: On Finding and Removing Artifacts in Language Data](https://mlsanity.com/paper/2104.08646/) (2021)
- [Inoculation by Fine-Tuning: A Method for Analyzing Challenge Datasets](https://mlsanity.com/paper/1904.02668/) (2019)

## Abstract (arXiv)

We study how informal surface forms degrade NLI accuracy in ELECTRA-small (14M) and RoBERTa-large (355M) across four transforms applied to SNLI and MultiNLI: slang substitution, emoji replacement, Gen-Z filler tokens, and their combination. Slang substitution (replacing formal words with informal equivalents, e.g., "going to" -> "gonna", "friend" -> "homie") causes minimal degradation (at most 1.1pp): slang vocabulary falls largely within WordPiece coverage, so the tokenizer handles it without signal loss. Emoji replaces content words with Unicode characters that ELECTRA's WordPiece tokenizer maps to [UNK], destroying the input signal before any learned parameters see it (93.6% of emoji examples contain at least one [UNK], mean 2.91 per example). Noise tokens (no cap, deadass, tbh) are fully in-vocabulary but absent from NLI training data, consistent with the model assigning them inferential weight they do not carry. The two failure modes respond to different interventions: preprocessing recovers emoji accuracy by normalizing text before tokenization; augmentation handles noise by exposing the model to noise-bearing examples during training. A hybrid of both achieves 88.93% on the combined variant for ELECTRA on SNLI (up from 75.88%), with no statistically significant drop on clean text. Against GPT-4o-mini zero-shot, unmitigated ELECTRA is significantly worse on transformed variants (p < 0.0001); hybrid ELECTRA surpasses it across all SNLI variants and reaches statistical parity on MultiNLI.