Emphasise encoder generation and continuous control; add measured sweep

README.md

@@ -2,22 +2,63 @@
 license: apache-2.0
 base_model: answerdotai/ModernBERT-large
 library_name: transformers
-pipeline_tag: fill-mask
+pipeline_tag: text2text-generation
+inference: false
 tags:
   - sentiment-control
+  - continuous-control
   - controllable-text-generation
+  - encoder-generation
+  - non-autoregressive
+  - masked-language-model
+  - text-style-transfer
+  - sentiment-transfer
   - text-rewriting
   - modernbert
+  - emnlp2026
 language:
   - en
 ---
 
 # SenseShift-large
 
+> **Generation from an encoder, with continuous control.**
+> An accepted EMNLP 2026 paper. Not a classifier, not a decoder — a masked LM
+> that *writes* text at a sentiment you dial in.
+
 SenseShift rewrites a sentence — or writes a new one — at **any sentiment you ask
 for on a continuous −1.0 to +1.0 scale**, while keeping it consistent with the
 surrounding text.
 
+Two things make it unusual:
+
+- **Continuous, not categorical.** Most sentiment-controlled generation offers
+  you `positive` / `negative` / `neutral`. SenseShift takes a *number* — 21
+  points on a 0.1 grid from −1.0 to +1.0. Sweeping the target over one sentence
+  of the example below, holding everything else fixed:
+
+  | target | achieved | rewritten sentence |
+  | ---: | ---: | --- |
+  | −1.0 | +0.0 | He said hello and took our order. |
+  | −0.9 | −0.8 | He looked sad and we felt bad. |
+  | −0.6 | −0.6 | He was very tired from working hard. |
+  | −0.3 | +0.0 | He said hello and took our order. |
+  | +0.0 | +0.0 | He served us some food and drinks. |
+  | +0.3 | +0.3 | We ordered a salad and shared it. |
+  | +0.6 | +0.5 | He smiled and gave us our drinks. |
+  | +0.9 | +0.8 | He smiled and thanked us for coming. |
+  | +1.0 | +0.8 | He smiled and thanked us for coming. |
+
+  The trend is monotone and most targets land within 0.1–0.2, but the control is
+  not exact: `−1.0` and `−0.3` both missed here, and the saturated ends (`±1.0`)
+  tend to collapse onto their `±0.9` neighbours. Treat the dial as a strong
+  steer, not a guarantee — and check `out.achieved_sentiment` if you need one.
+- **The generator is an encoder.** Text generation is nearly always
+  autoregressive decoding. SenseShift generates by iteratively filling masked
+  positions in a bidirectional ModernBERT, so every token it writes is
+  conditioned on the text on *both* sides. That is what lets it drop a
+  replacement sentence into the middle of a passage and have it fit.
+
 It is a masked language model fine-tuned from ModernBERT-large with an explicit
 **control vocabulary**: 21 special tokens `[-1.0] … [1.0]` on a 0.1 grid. At
 training time every sentence is prefixed with its own VADER sentiment token and
@@ -112,9 +153,15 @@ itself is a stock `ModernBertForMaskedLM` and can be loaded with
 
 - English only; trained on short narrative and review-style text.
 - VADER supplies the sentiment labels, so the model inherits its lexicon-based
-  view of sentiment. `achieved_sentiment` typically lands within ~0.2 of the
-  target rather than hitting it exactly.
+  view of sentiment. `achieved_sentiment` usually lands within 0.1–0.2 of the
+  target, but individual targets can miss outright — see the sweep above. The
+  saturated ends (`±1.0`) are the least reliable.
 - The rewrite is length-bounded by `num_masks`, so very long sentences are
   usually replaced by something shorter.
 - Because the whole passage is re-encoded per mask fill, generation cost grows
   with `num_masks × beam_size`.
+
+## Citation
+
+Accepted to **EMNLP 2026**. A preprint and the full BibTeX entry will be linked
+here once they are public.