aiqbench analysis
This research proposes the "interlingua hypothesis," suggesting that large language models perform translation by mapping source text into a shared, latent feature space before generating target text. The study provides convergent evidence through three lines of inquiry, demonstrating that translation capabilities emerge from monolingual competences rather than requiring specific language-pair interactions or aligned parallel data. This finding fundamentally shifts the understanding of LLM mechanics, indicating that robust translation can be achieved by leveraging general-purpose language modeling skills.
The Interlingua Hypothesis: LLMs Translate via a Latent Task-agnostic Feature Space
arXiv:2609.00515v1 Announce Type: cross Abstract: Large language models (LLMs) have recently demonstrated improved machine translation performance over strong supervised baselines. This raises questions as to what mechanisms underlie how LLMs perform machine translation between languages. Motivated by recent interpretability findings--namely, that LLMs use massively multilingual latent feature representations to perform language modeling--we propose the interlingua hypothesis. The hypothesis holds that language models translate by reading a source sentence into a latent feature space, and generate a target sentence by reading from the latent feature space. We show three lines of evidence in support of this hypothesis: (1) variance in BLEU across language pairs is largely predictable from language-specific competences with no language pair-specific interaction terms; (2) many model components are causally influential in both monolingual tasks and translation tasks; and (3) fine-tuning on monolingual data recovers a large proportion of translation improvements relative to fine-tuning on aligned documents. Together, these provide convergent evidence in support of the interlingua hypothesis, and suggest new ways of understanding and improving how LLMs can be leveraged to perform translation tasks.
Abstract:Large language models (LLMs) have recently demonstrated improved machine translation performance over strong supervised baselines. This raises questions as to what mechanisms underlie how LLMs perform machine translation between languages. Motivated by recent interpretability findings--namely, that LLMs use massively multilingual latent feature representations to perform language modeling--we propose the interlingua hypothesis. The hypothesis holds that language models translate by reading a source sentence into a latent feature space, and generate a target sentence by reading from the latent feature space. We show three lines of evidence in support of this hypothesis: (1) variance in BLEU across language pairs is largely predictable from language-specific competences with no language pair-specific interaction terms; (2) many model components are causally influential in both monolingual tasks and translation tasks; and (3) fine-tuning on monolingual data recovers a large proportion of translation improvements relative to fine-tuning on aligned documents. Together, these provide convergent evidence in support of the interlingua hypothesis, and suggest new ways of understanding and improving how LLMs can be leveraged to perform translation tasks.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2609.00515 [cs.CL]
(or arXiv:2609.00515v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.00515 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 269 words
Related coverage
[AINews] not much happened today
Latent Space · Sep 18, 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Reflections on unlearning and inoculation
LessWrong · Sep 20, 2026
Game-Theoretic Disempowerment: Loss of Control to Agencyless AI
LessWrong · Sep 27, 2026