patient - Story 102317 | Emotion Vectors in Gemma4-31B

The sound of the television was too low, a steady hum that didn't bother her. She knew that her partner, Mark, was practicing his Spanish, and she had discovered this when he'd left the TV on. Now he was in the next room, repeating phrases, the accent thick.

She was making tea. She poured herself a cup and carried it over to the window, where she sat, legs curled beneath her. She knew he would be there for a while.

The rhythmic repetition, the stumbles, the slow, deliberate delivery – it was all part of the process. She smiled, took a sip of tea, and closed her eyes.

Emotion: patient

Cluster: Passivity

PC1 (Valence): 2.71 Positive

PC2 (Disposition): 0.40

Role in Research

This story is one of 1,000 stories generated for the emotion patient. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 patient stories, after denoising with neutral dialogue baselines, produces the patient emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the patient vector is projected through the unembedding matrix.

Promoted:

`rest`	0.467
`eventually`	0.415
`leisurely`	0.369
`慢慢`	0.357
`a`	0.355

Suppressed:

`无法`	-0.432
`even`	-0.428
`인해`	-0.421
`無法`	-0.415
`😣`	-0.391