--- pretty_name: SenAffLex-Provo language: - en tags: - psycholinguistics - eye-tracking - naturalistic-reading - lexical-processing - lexical-frequency - word-predictability - affective-norms - valence - arousal - dominance - computational-linguistics - reproducible-research - research-software - linear-mixed-effects-models - bayesian-modeling --- # SenAffLex-Provo ## Integrated Sentiment, Affective, and Lexical Research Datasets and Computational Resources **Creator and Maintainer:** Mohit Gupta **Institution:** University of Illinois Urbana-Champaign **Research Discipline:** Psycholinguistics and Computational Linguistics **Repository Type:** Research datasets, computational software, and reproducibility resources **Initial Publication Year:** 2026 **Documentation Version:** 1.0 **Last Updated:** October 2026 **Repository DOI:** [10.57967/hf/11234](https://doi.org/10.57967/hf/11234) **Repository URL:** https://huggingface.co/datasets/27guptamohit/SenAffLex-Provo --- ## 1. Overview **SenAffLex-Provo** is an evolving computational research repository developed to support investigations of lexical, affective, and eventually sentiment-related influences on eye-movement behavior during naturalistic reading. The research infrastructure combines psycholinguistic datasets, computational preprocessing procedures, statistical analysis software, model-selection workflows, research results, and reproducibility resources. The original computational preprocessing, dataset integration, triangulation, and associated software were developed and implemented by **Mohit Gupta** at the University of Illinois Urbana-Champaign. The current research infrastructure integrates information from three established research resources: 1. **Provo Corpus** — Eye-tracking observations and contextual word predictability. 2. **SUBTLEX-US** — Lexical frequency information. 3. **Warriner et al. (2013)** — Valence, arousal, and dominance norms. The current data-preparation pipeline supports the construction of: - Master Provo. - Lexical Provo. - Affective Provo. Future development is planned to incorporate sentiment information through the **Gupta Lexical English Norms for Sentiment (GLENS)** dataset. **The planned sentiment integration is not part of the current released datasets.** The repository distinguishes original computational contributions from the underlying third-party research resources and their applicable intellectual property rights. --- ## 2. Research Objectives The SenAffLex-Provo research program aims to support reproducible investigations of language processing during naturalistic reading. Its objectives include: 1. Integrating established psycholinguistic research resources. 2. Developing reproducible computational preprocessing procedures. 3. Improving lexical matching through normalization and lemmatization. 4. Constructing integrated lexical and affective research datasets. 5. Evaluating lexical and affective influences on eye movements. 6. Supporting statistical analyses of reading behavior. 7. Documenting computational methods and model-selection procedures. 8. Preserving research scripts, results, and computational artifacts. 9. Supporting independent reproduction and methodological extensions. 10. Expanding the infrastructure to incorporate sentiment information. These objectives describe the scope of the research program. Planned developments are distinguished from completed research resources. --- ## 3. Original Research Data Sources The current research infrastructure integrates three established psycholinguistic resources. | Source | Original Authors | Research Contribution | |---|---|---| | Provo Corpus | Luke & Christianson (2018) | Eye-tracking observations and contextual predictability | | SUBTLEX-US | Brysbaert & New (2009) | Lexical frequency norms | | Warriner Affective Norms | Warriner, Kuperman, & Brysbaert (2013) | Valence, arousal, and dominance ratings | The original source datasets remain attributable to their respective creators. Their inclusion in the SenAffLex-Provo computational workflow does not transfer ownership of the underlying observations, measurements, or other protected materials. For complete source citations and licensing considerations, see: [Raw Data Citations](Raw-Data-Citations.md) --- ## 4. Integrated Research Datasets The computational preparation pipeline supports three primary research datasets. | Dataset | Description | Research Application | |---|---|---| | Master Provo | Integrated Provo dataset retaining observations regardless of lexical or affective norm availability. | Corpus research, coverage analysis, and construction of analytical subsets | | Lexical Provo | Predictor-complete dataset containing mandatory lexical predictors. | Lexical Provo research program | | Affective Provo | Predictor-complete dataset containing mandatory lexical and affective predictors. | Affective Provo research program | ### 4.1 Master Provo Master Provo integrates available eye-tracking, predictability, lexical frequency, and affective information. It preserves the original token-level observation structure during lexical integration. ### 4.2 Lexical Provo Lexical Provo contains observations satisfying the required lexical predictor-completeness criteria. Its mandatory lexical predictors include: - Word length. - Contextual word predictability. - Lexical frequency. ### 4.3 Affective Provo Affective Provo contains observations satisfying the required lexical and affective predictor-completeness criteria. Its mandatory affective predictors include: - Valence. - Arousal. - Dominance. ### 4.4 Outcome-Specific Filtering The exported datasets are not automatically the final analytical samples for every statistical model. Outcome-specific filtering must be performed according to the requirements of each dependent variable. For example: - Fixation-duration analyses may require positive duration measurements. - Skipping analyses must retain skipped and non-skipped observations. - Regression analyses require appropriate outcome-specific eligibility criteria. Researchers should report the actual observation counts used in their statistical models. --- ## 5. Computational Research Infrastructure The repository contains original computational software developed for preparing, integrating, validating, and analyzing psycholinguistic research data. ### 5.1 Data Preparation The preprocessing pipeline includes: 1. Source-data preparation. 2. Provo Corpus processing. 3. Cloze predictability preparation. 4. Lexical normalization. 5. Vocabulary extraction and matching. 6. Lemmatization-related processing. 7. SUBTLEX-US frequency integration. 8. Warriner affective norm integration. 9. Dataset triangulation. 10. Predictor-completeness evaluation. 11. Dataset integrity validation. 12. Research dataset export. ### 5.2 Statistical Research The research infrastructure also contains computational resources supporting statistical investigations of eye-movement behavior. These include, where available: - Linear mixed-effects modeling. - Random-effects structure selection. - Model diagnostics. - Bayesian statistical analyses. - Computational sensitivity analyses. - Statistical results. - Reproducibility artifacts. The availability and scope of specific analyses are documented within their corresponding research directories. --- ## 6. Repository Organization ```text SenAffLex-Provo/ │ ├── README.md ├── LICENSE.md ├── CITATION.cff ├── SenAffLex-Provo-Data-Citation.md ├── Raw-Data-Citations.md ├── Research-Programs-Citations.md │ ├── 01-README/ │ └── Research documentation │ ├── 02-Data/ │ ├── 01-Raw/ │ ├── 02-Intermediate/ │ └── 03-Analysis_Ready/ │ ├── 03-Data-Preparation/ │ └── Computational preprocessing software │ ├── 04-Research_Program-Lexical_Provo/ │ └── Lexical research resources │ └── 05-Research_Program-Affective_Provo/ └── Affective research resources ``` The directory structure separates source data, processed research resources, computational preparation, and scientific investigations. Additional research directories may be introduced as the program expands. --- ## 7. Computational Development and Attribution The original computational preparation and triangulation infrastructure was developed and implemented by **Mohit Gupta**. Original contributions include, where applicable: - Python preprocessing implementation. - R statistical programming. - Jupyter notebook development. - Lexical normalization and lemmatization. - Vocabulary matching. - Dataset integration and triangulation. - Construction of intermediate datasets. - Construction of analysis-ready datasets. - Computational validation. - Statistical modeling workflows. - Research software documentation. - Reproducibility procedures. These statements identify responsibility for original computational implementation. They do not imply ownership of the original third-party research datasets or sole authorship of associated scientific publications. --- ## 8. Reproducibility The repository provides computational resources intended to support reproduction of the research workflows. Researchers should consult the relevant directory documentation before executing scripts or notebooks. Reproducibility may require: - Compatible Python and R environments. - Appropriate software dependencies. - Access to original source datasets. - Correct input and output directory configurations. - Compatible lexical processing resources. - Sufficient computing resources. - Consistent source-data versions. ### 8.1 Data Preparation [Data Preparation Documentation](03-Data-Preparation/README.md) ### 8.2 Analysis-Ready Datasets [Analysis-Ready Dataset Documentation](02-Data/03-Analysis_Ready/README.md) ### 8.3 Research Programs Research-specific computational requirements and analysis procedures are documented within their respective directories. Researchers should identify the repository version or Git commit corresponding to the resources used. --- ## 9. Licensing **SenAffLex-Provo is a mixed-license research repository.** Different licensing arrangements apply to different categories of materials. | Material | Licensing Arrangement | |---|---| | Original Python software | MIT, where designated | | Original R software | MIT, where designated | | Original Jupyter notebook code | MIT, where designated | | Eligible original documentation | CC BY 4.0, where designated | | Eligible original dataset contributions | CC BY 4.0, where designated | | Original Provo Corpus | Applicable original source terms | | SUBTLEX-US | Applicable original source terms | | Warriner affective norms | Applicable original source terms | | Third-party software dependencies | Applicable original software licenses | ### 9.1 Software Licensing Eligible original computational software is distributed under the MIT License where designated. The MIT License permits use, modification, redistribution, and commercial reuse subject to preservation of applicable copyright and permission notices. ### 9.2 Dataset Licensing The complete integrated datasets may contain information governed by third-party rights. **The repository does not independently offer unrestricted CC BY 4.0 permissions for the complete contents of the integrated datasets.** ### 9.3 Intermediate Dataset Licensing The intermediate-data licensing documentation distinguishes four preprocessing outputs. Files 1 and 2 were processed by Mohit Gupta but are excluded from the intermediate-data CC BY 4.0 licensing grant. Eligible original contributions in Files 3 and 4 are offered under CC BY 4.0, subject to applicable third-party rights. ### 9.4 Third-Party Redistribution Public availability of a dataset does not automatically establish unrestricted redistribution permission. The applicable redistribution permissions for the original Provo Corpus, SUBTLEX-US, and Warriner affective norms remain under review. For complete information: [Repository Licensing Policy](LICENSE.md) --- ## 10. How to Cite SenAffLex-Provo The repository distinguishes three categories of scholarly citation: 1. Original source datasets. 2. SenAffLex-Provo computational resources. 3. Associated scientific publications. Researchers should cite the resources relevant to their work. ### 10.1 Official Repository Citation Researchers substantially using the original SenAffLex-Provo computational resources are strongly requested to cite: Gupta, M. (2026). *SenAffLex-Provo: Integrated sentiment, affective, and lexical research datasets and computational resources* [Data set and software]. Hugging Face. https://doi.org/10.57967/hf/11234 ### 10.2 Original Source Citations Researchers using the original datasets should cite their respective publications. [Raw Data Citations](Raw-Data-Citations.md) ### 10.3 Associated Scientific Publications Researchers using scientific findings or interpretations from an associated publication should cite that specific article. [Research Programs and Publications](Research-Programs-Citations.md) ### 10.4 Machine-Readable Citation The official machine-readable repository citation is provided in: [CITATION.cff](CITATION.cff) ### 10.5 Citation Guidance For detailed guidance on citing the integrated datasets and computational resources: [SenAffLex-Provo Data Citation](SenAffLex-Provo-Data-Citation.md) Formal academic citation of the repository is strongly requested but is not an additional legal condition of the standard MIT License. --- ## 11. Associated Research Publications The SenAffLex-Provo research infrastructure supports multiple scientific investigations. ### Affective Provo — Fixation Duration **VAD Paper 1** Gupta, M., & Christianson, K. (in press). Beyond word length, frequency, and predictability: Affective properties predict eye movement in reading. *Reading Research Quarterly*. **Status:** Accepted October 20, 2026. **Article DOI:** Pending assignment. For the authoritative and continuously updated publication bibliography, see: [Research Programs Citations](Research-Programs-Citations.md) --- ## 12. Future Development The SenAffLex-Provo research infrastructure is intended to support continued scientific development. Planned extensions include: - Integration of GLENS sentiment norms. - Additional lexical and affective analyses. - Fixation-behavior investigations. - Regression-behavior investigations. - Expanded statistical modeling resources. - Additional scientific publications. - New reproducibility materials. Planned resources are not represented as completed or published. --- ## 13. Repository Versioning The repository may evolve as additional datasets, scripts, research results, and documentation become available. Researchers are encouraged to record: 1. Repository DOI. 2. Dataset or software resource used. 3. Relevant Git commit or release identifier. 4. Date of access, where appropriate. 5. Modifications made to the original resources. The DOI identifies the repository resource, while a Git commit identifies a particular repository state. --- ## 14. Contact **Repository Creator and Maintainer:** Mohit Gupta **Institution:** University of Illinois Urbana-Champaign **Repository:** https://huggingface.co/datasets/27guptamohit/SenAffLex-Provo For questions concerning computational resources, licensing documentation, or reproducibility, please use the repository's Hugging Face discussion facilities. --- ## 15. Important Notice The SenAffLex-Provo repository contains original computational contributions alongside information originating from independently developed research resources. The original Provo Corpus, SUBTLEX-US, and Warriner affective norms remain attributable to their respective creators. The repository's MIT and CC BY 4.0 licensing arrangements do not override third-party intellectual property rights. Researchers must comply with applicable original licensing and redistribution conditions. The repository DOI does not independently establish permission to redistribute third-party research data.