SMCLM: Semantically Meaningful Causal Language Modeling for Autoregressive Paraphrase Generation

MICHAŁ PEREŁKIEWICZ, SŁAWOMIR DADAS, AND RAFAŁ POŚWIATA · 2025

Read SMCLM: Semantically Meaningful Causal Language Modeling for Autoregressive Paraphrase Generation by MICHAŁ PEREŁKIEWICZ, SŁAWOMIR DADAS, AND RAFAŁ POŚWIATA (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03415.txt.

Description

Michał Perełkiewicz, Sławomir Dadas, and Rafał Poświata teach language models to restate text in new words while keeping the meaning intact. Their paper appeared in two thousand twenty-five. Paraphrase tools help with writing, translation checks, and data growth. Most top tools need labeled pairs, which are costly to build.

Their method starts training from a meaning-rich embedding of the source text. That anchor keeps the model's output tied to the sense, not just the surface. The self-taught scheme matched supervised rivals and topped other unsupervised ones. The team also built a broad set of scores for judging paraphrases. In doing so they show that the field's favorite scores are unreliable. Their metric suite gives future work a firmer yardstick.

Catalog record

CREATORMICHAŁ PEREŁKIEWICZ, SŁAWOMIR DADAS, AND RAFAŁ POŚWIATA
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-03415

Browse the full library · About the Tower of Babel