SENSESHIFT: Continuous Sentiment-Controlled Text Generation via Encoder-based Mask Infilling

Shahed Masoudian; Markus Frohmann; Emmanouil Karystinaios; Navid Rekabsaz; Markus Schedl

cs.CLAug 25, 2026

SENSESHIFT: Continuous Sentiment-Controlled Text Generation via Encoder-based Mask Infilling

Authors: Shahed Masoudian, Markus Frohmann, Emmanouil Karystinaios, Navid Rekabsaz, Markus Schedl

Organizations: Johannes Kepler University Linz · Thomson Reuters Labs · University of Toronto, Vector Institute

Abstract

Recent controllable text generation (CTG) for sentiment control has largely focused on decoder-based large language models, making causal attention the dominant paradigm. While effective for fluent generation, these models still struggle to satisfy complex constraints and follow fine-grained sentiment signals specified by users. Existing sentiment-aware CTG methods typically simplify the problem by treating sentiment either as a coarse categorical label (e.g., positive or negative) or as a single fine-grained control signal applied to an entire document. Consequently, more challenging settings such as sentence-level sentiment control within long-form text remain underexplored. To address these limitations, we introduce SenseShift , an encoder-based framework for fine-grained sentence-level CTG. Unlike standard decoder architectures, SenseShift leverages bidirectional attention, quantized sentiment signals, and iterative mask infilling to generate local sentences conditioned on target sentiment intensity. Empirical evaluations on story and review generation demonstrate that SenseShift achieves stronger sentiment controllability while maintaining text quality and robustness to out-of-domain generation compared to larger decoder-based baselines.

Figures & tables

Figure 1: SenseShift VS decoder-based prompting for sentiment CTG. Based on a given text, target sentence and sentiment, SenseShift places sentiment tokens before each sentence. encoder-architecture allows SenseShift to predict masked tokens attending to each target sentiment while distinguishing prior context from posterior.

Model	Story Evaluation						Review Evaluation
Model	PPL old	PPL ${}_{new}\downarrow$	$\Delta_{s}$ $\downarrow$	$\Delta_{f}$ $\uparrow$	Corr. $\uparrow$	Acc. $\uparrow$	PPL old	PPL ${}_{new}\downarrow$	$\Delta_{s}$ $\downarrow$	$\Delta_{f}$ $\uparrow$	Corr. $\uparrow$	Acc. $\uparrow$
Prompting
ModernBert-D-28m	80.6	255.5	0.56	-0.29*	0.07	0.36	222.3	161.2	0.69	-0.13*	-0.19	0.23
Llamma3.2-3b	81.1	118.1	0.50	-0.05*	0.25	0.41	239.4	113.1	0.51	-0.00	0.21	0.38
Gemma2-2b	88.6	139.2	0.56	-0.00	0.15	0.37	245.1	117.3	0.55	0.02*	0.14	0.37
Gemma2-9b	83.6	126.2	0.39	-0.03*	0.65	0.55	268.3	97.8	0.44	0.00	0.53	0.50

Table 1: Overall comparison between baselines and SenseShift . Significantly different changes (

p

-value<0.01) compared to the 0 baseline are marked with an asterisk. Note that columns under green are same models trained on story and columns under red are same models trained on review. in table

PPL_{old}

refers to perplexity of the original sentence calculated using GPT-2 and

PPL_{new}

refers to generated sentence perplexity

Figure 2:

\Delta_{s}

across three evaluation dimensions. Shaded regions indicate variance. Left:

\Delta_{s}

vs. target sentiment. Center:

\Delta_{s}

vs. relative sentence position. Right:

\Delta_{s}

vs. number of masked tokens.

Criterion	Gemma2-2b	SenseShift	Both	Neither
Fitness	22.1	23.0	46.9	8.0
Fluency	6.2	6.2	85.0	2.7
Sentiment	28.3	31.0	28.3	12.4
Overall	18.9	20.1	53.4	7.7

Table 2: Users preference (

\%

) of sentences generated by ModernBert-E-0.4b -story vs Gemma2-2b trained with Token-Instruction-Tuning. Rows correspond to evaluation criteria: contextual fitness, grammatical correctness, and sentiment following.

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 3: Sentiment of story ending written by GPT4o-Mini for Negative, Positive, and Neutral ending with and without intensifying adverb "Very" in the prompt.

Text
Statistics	TinyStories	Yelp
Num Samples	4439352	252085
Positive (>0.2)	53.7%	41.0%
Negative (<-0.2)	0.3%	4.3%
Neutral	46.0%	54.8%
Avg Sentences	14.2 ± 7.77	6.4 ± 4.86

Appendix

Table 3: Dataset statistics for the train splits. Sentiment scores are in

[-1,1]

State	Hyperparameter	Prompt/Steer	Inst-Tune	Token-Inst-Tune	SenseShift
Training	Learning rate (lr)	–	1e-5	1e-5	1e-5
	Batch size (bs)	–	32	32	50
	Dropout (dp)	–	0.0	0.0	0.0
	Warmup steps	–	500	500	500
	Epochs	–	30	30	50
	Weight decay	–	0.01	0.01	0.01

Appendix

Table 4: Default hyperparameters used across different training paradigms for controllability. * value for Qwen is 1024 to ensure finalizing thinking process.

Model	Train	Story Generation					Review Generation
Model	Train	Ppl. $\downarrow$	$\Delta_{s}$ $\downarrow$	$\Delta_{f}$ $\uparrow$	Corr. $\uparrow$	Acc. $\uparrow$	Ppl. $\downarrow$	$\Delta_{s}$ $\downarrow$	$\Delta_{f}$ $\uparrow$	Corr. $\uparrow$	Acc. $\uparrow$
Whole Mask Prediction
ModernBert-E-0.15b	Story	596.8	0.42	-0.01	0.54	0.44	303.1	0.44	-0.02*	0.48	0.38
ModernBert-E-0.15b	Review	482.9	0.50	-0.04*	0.27	0.31	325.2	0.43	-0.01	0.55	0.44
ModernBert-E-0.4b	Story	869.9	0.4	-0.039*	0.55	0.5	571.3	0.43	0.01	0.48	0.39
ModernBert-E-0.4b	Review	426.3	0.50	-0.041*	0.29	0.32	288.8	0.44	-0.013	0.51	0.41

Appendix

Table 5: Comparison between Iterative Mask Infilling and Whole Mask prediction

Criterion	Gemma2-2b	SenseShift	Both	Neither
Fitness	17.2	23.4	54.7	4.7
Grammar	4.7	4.7	89.1	1.6
Sentiment	29.7	28.1	31.2	10.9
Overall	17.2	18.8	58.3	5.7

Appendix

Table 6: Human preference evaluation on the Story domain (n=55 sentence pairs). The specific model used for evaluation is ModernBert-E-0.4b trained on stories. Similarity to Original is reported as a diagnostic metric and excluded from the Overall score.

Criterion	Gemma2-2b	SenseShift	Both	Neither
Fitness	27.7	23.4	36.2	12.8
Grammar	8.5	8.5	78.7	4.3
Sentiment	25.5	36.2	23.4	14.9
Overall	20.6	22.7	46.1	10.6

Appendix

Table 7: Human preference evaluation on the Review domain (OOD n=45 sentence pairs). The specific model used for evaluation is ModernBert-E-0.4b trained on stories. Similarity to Original is reported as a diagnostic metric and excluded from the Overall score.

Figure 4: Mean $\Delta_{s}$ as a joint function for the ModernBert-E-0.4b -story. Left: Target sentiment vs. relative sentence position. Center: Target sentiment vs. number of masked tokens. Right: Relative sentence position vs. number of masked tokens. Lower values (green) indicate better sentiment control.

Figure 5: Interactive demo of SenseShift . Users can choose or randomly select data from both dataset and edit different sentences with their desired sentiment intensity

Panel of Figure 6 — Figure 6: Qualitative examples of iterative sentiment editing on a story (in-distribution). Each subfigure shows the sentiment arc (top) and annotated text (bottom).

Panel of Figure 7 — Figure 7: Qualitative examples of iterative sentiment editing on a hotel review (out-of-distribution). Each subfigure shows the sentiment arc (top) and annotated text (bottom).

Controllable Text Generation

Controllable text generation steers language models toward desired attributes, formats, audiences, and constraints while preserving useful, natural text.

See all 109 papers

LiSeCo: Linear Semantic Control for Language Generation
May 24, 2024Emily Cheng, Carmen Amo AlonsoLanguage Model Steering Controllable Text Generation
Universitat Pompeu Fabra, Barcelona, Spain · Stanford University, CA, USA
Steering Language Models Before They Speak: Logit-Level Interventions
Jan 16, 2026Hyeseon An, Shinwoo Park, Hyundong Jin +1Language Model Steering Controllable Text Generation
Department of Computer Science, Yonsei University, Seoul, South Korea
Post-Training Zero-Shot TTS for Fine-Grained Emotion and Duration Control via Natural Language
Sep 10, 2026Lianru Gao, Yujie Guo, Yong QinTTS Synthesis Zero-Shot TTS
TMCC, College of Computer Science, Nankai University, Tianjin, China

May 24, 2024J/K move · Enter open · S save

LiSeCo: Linear Semantic Control for Language Generation

Emily Cheng, Carmen Amo Alonso

Universitat Pompeu Fabra, Barcelona, Spain · Stanford University, CA, USA

Language Model Steering Controllable Text Generation LLM Safety Activation Steering

Open paper AskPDF arXiv

The prevalence of Large Language Models (LLMs) in critical applications highlights the need for controlled language generation methods that are both computationally efficient and enjoy performance guarantees. To address this need, we use a common model of concept semantics as linearly represented in an LLM's latent space. In particular, we take the view that natural language generation traces a trajectory in this continuous semantic space, realized by the language model's hidden activations. This view permits a control-theoretic treatment of text generation in latent space, in which we propose Linear Semantic Control (LiSeCo), a lightweight, gradient-free intervention that dynamically steers trajectories away from regions corresponding to undesired meanings. In particular, we propose to directly intervene, in an online fashion, the activations of the token that is being generated in embedding space. Crucially, LiSeCo does not simply steer activations towards a desirable region. Instead, it relies on classical techniques from control theory to precisely control activations in a context-dependent way, and guarantees that they are brought into a specific pre-defined region of embedding space that corresponds to allowed semantics. The intervention is computed in closed form according to an optimal controller formulation, minimally impacting generation time. This control of the activations in embedding space allows for fine-grained steering of attributes of the generated sequence. We demonstrate that our approach is effective on different tasks -- toxicity, sentiment, and language (English/Spanish) steering -- while maintaining text quality.

Abstract

Controllable Text Generation

Explore similar work

LiSeCo: Linear Semantic Control for Language Generation

Steering Language Models Before They Speak: Logit-Level Interventions

Post-Training Zero-Shot TTS for Fine-Grained Emotion and Duration Control via Natural Language