# Text Degeneration

> Text degeneration is the tendency of language models, especially under greedy or beam-search decoding, to produce bland, repetitive, or looping text, such as the same phrase repeated over and over.

- Identifier: PTL-0175
- Category: Failure Modes & Evaluation
- Canonical URL: https://protologue.com/t/text-degeneration/
- Also known as: neural text degeneration, repetition, repetition loops
- Introduced: 2019

## Description

Maximizing likelihood at each step favors safe, high-probability continuations that reinforce themselves once repetition starts. Sampling methods such as top-p and top-k, repetition or frequency penalties, and modern training largely mitigate it, but loops still appear in long generations and agent traces.

## Related terms

- [Greedy Decoding](https://protologue.com/t/greedy-decoding/)
- [Top-p Sampling](https://protologue.com/t/top-p-sampling/)
- [Top-k Sampling](https://protologue.com/t/top-k-sampling/)
- [Temperature](https://protologue.com/t/temperature/)

## Sources

- Holtzman et al. (2019). The Curious Case of Neural Text Degeneration. https://arxiv.org/abs/1904.09751

## Cite this entry

Protologue. (2026). Text Degeneration. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0175). https://protologue.com/t/text-degeneration/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)