Ariel University · Department of Computer Science

NLP for Human Sciences Lab

Bridging the gap between Large Language Models and the human experience — applying computational linguistics to deepen scientific understanding of mind, society, and culture.

Meet the Team Research Lines Active Projects

What We Do

We focus on the development and application of advanced tools from natural language processing to deepen the scientific understanding of people, society, and culture. As an interdisciplinary hub, we position Large Language Models as both practical instruments and testable scientific environments for studying cognition, culture, and clinical interaction.

We investigate the use of large-scale language models to reveal new insights in human-centered sciences — particularly within clinical psychology and social dynamics — transforming complex textual data, from therapy transcripts to social media, into a window through which we can observe and quantify human behavior and mental states.

In parallel, we analyze the behavior and internal representations of language models themselves to better understand the nature of language, cognition, and the relationship between artificial intelligence and the human experience — through mechanistic interpretability and latent space analysis.

Three Research Lines

01

Computational Clinical Psychology

Transforming clinical interactions into quantifiable data using NLP. We analyze therapist-patient dynamics to map mechanisms of psychological change and identify digital biomarkers of mental states — providing clinicians with empirical, data-driven insights rather than subjective observations alone.

Mapping the mechanisms of psychological change through therapist-patient dynamics and session-level outcome prediction.

02

NLP as Scientific Accelerator

Serving researchers across medical sciences and digital humanities by bridging raw textual data and scientific insight. We build NLP workflows that allow domain scientists to analyze large text corpora with the depth and speed needed for rigorous empirical discovery.

Enabling researchers to ask questions at a scale and resolution previously not feasible.

03

Mechanistic Interpretability & In-Silico Science

Examining AI's internal structures as experimental environments. We identify conceptual dimensions of latent representations corresponding to human constructs, using model perturbations for causal simulation — advancing scientific methodology through active experimentation within the learned structure of language models.

Turning LLMs into in-silico laboratories for theory-driven cognitive and social science experiments.

Active Research Projects

Clinical NLP

Sequences that Heal: Discovering Effective Patterns in Psychotherapy

Using annotated psychotherapy data to identify which sequences of therapist interventions and patient self-states foster positive change, connecting NLP-detected patterns to treatment outcomes.

Collab: Prof. Dana Atzil-Slonim (Bar-Ilan University)

Clinical NLP

Quantifying the Unspoken: Computational Modeling of Mentalization in Therapy

Developing NLP frameworks to automatically identify mentalization (reflective functioning) in clinical transcripts, examining how mentalizing-focused techniques facilitate psychological growth.

Collab: Prof. Mario Mikulincer (Hebrew University of Jerusalem)

Clinical NLP

Automating MIND Prediction: Micro-level Outcomes & Dyadic Dynamics

Using language models to predict mental states at fine-grained segment-level intervals, focusing on linguistic synchrony and emotional exchange to track immediate shifts in session quality within five-minute windows.

Interpretability

Probe and Perturb: In-Silico Experiments via Latent Space Interventions

The lab's flagship project. Probing LLM latent spaces to identify directions aligned with constructs like therapist directiveness and patient defensiveness, enabling controlled in-silico experiments.

Interpretability

LoCAV: Identifying Low-Rank Conceptual Vectors in Latent Space

Developing methodologies to isolate conceptual vectors within LLMs using low-rank subspaces and concept activation vectors, supporting white-box AI transparency and targeted model interventions.

Language & Register

Multilingual Slang and Register Representations

Investigating how multilingual language models encode different linguistic registers — particularly informal language and slang — across diverse language families.

arXiv  ·  OpenReview

Digital Humanities

Kabbalah and Cross-Cultural Textual Influence

Applying computational methods to analyze cross-cultural textual influence and transmission patterns in historical and religious texts, exemplifying the lab's NLP-as-accelerator vision in the humanities.

Join the Lab

For Students

We are looking for motivated researchers interested in the intersection of LLMs, psychology, and interpretability. If you want to work at the frontier of AI and human science, reach out with a brief description of your background and interests.

ayal.s.klein@gmail.com

For Collaborators

We are always open to new interdisciplinary partnerships where language-related technologies can unlock new perspectives in scientific or clinical textual data. If you have a research question and a corpus, let's talk.

ayal.s.klein@gmail.com