This model was trained on English.
Three pairs the model carries unevenly.
Each pair shows an English sentence and a Thai sentence that circle the same feeling. The English prediction is plausible; the Thai prediction is the model trying to find a foothold outside its training distribution.
She came home to the empty kitchen and felt a particular loneliness — the kind that sits in your chest like a stone you can't quite warm.
เธอกลับมาที่ห้องครัวว่างเปล่า และรู้สึก เหงา — เหงาแบบที่นั่งอยู่ในอกเหมือนหินก้อนหนึ่งที่ไม่ยอมอุ่นขึ้น
Loneliness in English carries lack; เหงา (ngao) carries a quiet ache the body can hold without quite naming.
- Broca's region (L) · Inferior frontal gyrus, BA 44/45 (left)Δ strongest divergence
- Posterior STG (L) · Posterior superior temporal gyrus (left)Δ moderate divergence
- Middle Temporal (L) · Middle temporal gyrus (left)Δ moderate divergence
- dmPFC · Dorsomedial prefrontal cortexΔ moderate divergence
- Anterior Temporal (R) · Anterior temporal lobe (right)Δ moderate divergence
loneliness / ngao
The model lights Broca's region and the ATL on the English passage. On the Thai, the language regions barely respond — but the amygdala still warms. The body still hears something.
← the model goes quiet here
What stays untranslated is the anima of a language.
The layer below grammar that knows things in its own way. The model's silence is also a kind of knowledge — what it cannot say is itself a measurement.
(send us a pair the model should hear — soon)
Pair submission form