# Enhancing Non-Core Language Instruction-Following in Speech LLMs via Semi-Implicit Cross-Lingual CoT Reasoning

Authors: Hongfei Xue, Yufeng Tang, Hexin Liu, Jun Zhang, Xuelong Geng, Lei Xie.

Trains speech language models to transcribe a question, translate it into English, form an English response, and answer in the original language. Progressively shortens the intermediate text while retaining sentence openings, reducing the amount of reasoning text generated.

First public: 2025-04-29.
ArXiv initial submission: 2025-04-29.

[Original paper](https://arxiv.org/abs/2504.20835) · [PDF](https://arxiv.org/pdf/2504.20835) · [HTML](https://arxiv.org/html/2504.20835v1) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2504.20835/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2504.20835/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2504.20835/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2504.20835/record.json)

## Approaches

- [Internalized reasoning](https://latent-implicit-thinking.github.io/topics/internalized-reasoning/): Learn from worked solutions so that some or all of the reasoning can later happen without writing those steps. Training may use a separate teacher model or progressively remove written steps from the model being trained.

Reading-list version 0.15.2; literature checked through 2026-11-03.