Mix and Match: Context Pairing for Scalable Topic-Controlled Educational Summarisation
Nathikan Yodthapa, Thanapong Intharah, Sahan Bulathwela · 2026
Read Mix and Match: Context Pairing for Scalable Topic-Controlled Educational Summarisation by Nathikan Yodthapa, Thanapong Intharah, Sahan Bulathwela (2026) 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-2604-18087.txt.
Description
Yodthapa and two colleagues study data augmentation for topic-controlled summarization with small language models, which must focus summaries on specific aspects of source texts. Their pairwise method mixes contexts from other texts into contrastive examples, teaching models the link between topics and summaries better. Using SciTLDR data with Wikipedia topics, they show win rate and semantic match rise as augmentation scale grows while real data stays fixed. A T5-base model trained this way rivals larger models despite far fewer parameters and real examples.
Catalog record
| CREATOR | Nathikan Yodthapa, Thanapong Intharah, Sahan Bulathwela |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19587 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.18087