The frozen BETO-2019 design and the modern e5-base encoder are numerically close on Spanish, with no descriptive separation on either task¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? On the Spanish held-out set, frozen BETO-2019 embeddings and the modern multilingual e5-base encoder show no descriptive separation on thematic or sentiment classification. → Esteban U. Vega Barajas 2026

Evidence¶

Esteban U. Vega Barajas 2026¶

Esteban U. Vega Barajas. (2026). A Durability and Cross-Language Transfer Benchmark for a Validated Teaching-Feedback Classification Protocol. https://arxiv.org/abs/2607.11873 doi:10.48550/arxiv.2607.11873

q2 · i? · causal · r2

Descriptive paired comparison of the two frozen-embedding arms under the identical protocol on the 233-comment Spanish held-out split. The article reports near-even McNemar counts (thematic 𝑏=29, 𝑐=28; sentiment 𝑏=12, 𝑐=17) and states "no significance test is run".

"Frozen BETO-2019 and the modern e5-base encoder are numerically close, with no descriptive separation: thematic McNemar 𝑏=29, 𝑐=28 and sentiment 𝑏=12, 𝑐=17, and no significance test is run."

Discussion¶

Related Claims¶

• Sentiment classification transfers to English under the unchanged protocol, with a frozen multilingual encoder the strongest arm at wF1 0.73 and LLMs matching it only with few-shot prompting — related
• On Spanish sentiment, the 2026 frontier model shows no advantage over the cheap model, with Haiku zero-shot at 0.923 versus Opus zero-shot at 0.884 and the head-to-head McNemar favoring Haiku — related
• Under the fixed protocol, hosted LLMs produce the largest gain on Spanish thematic classification, with thematic macro F1 rising from 0.46 (TF-IDF) to roughly 0.82-0.86 for the two hosted LLMs — related