# Crow Intelligence

> An independent research lab studying how language shapes thought. We combine cognitive linguistics with computational methods — NLP, corpus analysis, network science — to examine how people and institutions actually reason, beneath the official version.

Crow Intelligence is run by Zoltán Varjú (Founder & AI Advisor) and Orsolya Putz, PhD (Partner & Cognitive Scientist), based in Hungary and working across the EU. Its primary output is long-form visual essays and interactive research projects, published openly, together with the open-source Python packages they are built on. The lab also takes on a limited amount of commissioned research and AI advisory work.

Two things are worth knowing when reading this site by machine. First, the essays and projects are scrollytelling pages: the argument is carried by visualisations as much as by prose, so the rendered HTML understates them — the summaries below state what each one does and what it found. Second, some work is published in Hungarian; that is marked.

This file follows the llms.txt convention (https://llmstxt.org/).

## Aporia — visual essays

Aporia is the lab's essay series, and the best starting point for understanding the work. Each essay is a self-contained piece of original research: it states its own method, shows its data, and argues a case. Titles are drawn from the text under discussion.

- [The Wrath and the Journey](https://crowintelligence.org/aporia/homer/): Compares the Iliad and the Odyssey as corpora. Extracts proper nouns and themes, builds a character co-occurrence network (the "small world" of Greek heroes), and maps the geography of each poem — finding that 21.3% of the Odyssey is open water, and reading the Catalogue of Ships against the wanderings.
- [The great modern crimes are public crimes](https://crowintelligence.org/aporia/nagel-essay/): Measures the gap between public and private personality in American leaders by scoring Big Five traits on public speeches versus private letters and diaries. Lincoln reads as the same man in both registers; Nixon does not. Takes its title and its question from Thomas Nagel's essay on ruthlessness in public life.
- [Whereof one cannot speak, thereof one must be silent](https://crowintelligence.org/aporia/wittgenstein/): Reads Wittgenstein's Tractatus as a shape: the whole text walked out as one continuous right-angle spiral through its numbered propositions, set in handwriting and coloured by topical cluster. Companion visualisations lift the spiral into 3-D and strip the text to its punctuation alone.
- [Where does the mind end and the world begin?](https://crowintelligence.org/aporia/epistemic-arcade/): An empirical test of Clark & Chalmers' extended-mind thesis. Two reinforcement-learning agents (PPO) learn Tetris under identical architectures but opposite reward functions — one charged for every keystroke, one free to rotate pieces on screen rather than in its head. Their strategies diverge as the epistemic actions are priced in or out.
- [Nekem írod a dalt](https://crowintelligence.org/aporia/magyar-dalszovegek-essze/): Traces seven words — szerelem (love), lány (girl), csaj (chick), pénz (money), éjszaka (night), szabadság (freedom), élet (life) — through seventy years of Hungarian popular song, quoting real lyrics decade by decade and tracking how each word shifts meaning and company. Written in Hungarian.

## Interactive research projects

Standalone dashboards and visualisations, each published as its own microsite.

- [The Wealth of Nations — 250th Anniversary Digital Edition](https://crowintelligence.org/site/index.html): An interactive exploration of Adam Smith's 1776 masterwork. Topic modeling, named entity recognition, rhetorical metaphor analysis, and full-text search across all five books.
- [The Nagel Index — Public vs. Private Personality](https://crowintelligence.org/aporia/nagel-essay/): Big Five personality analysis of political leaders, measuring the gap between their public speeches and private correspondence. Inspired by Thomas Nagel's essay on ruthlessness in public life.
- [Market Metaphors — Kindleberger Phases × Narrative Economics](https://crowintelligence.org/web/index.html): Tracing the evolution of metaphorical language in 158,666 financial news headlines (2009–2020) across five crisis events, mapping metaphor domains to the Minsky–Kindleberger panic cycle.
- [The Narrative Engine — FOMC Speech Analysis](https://crowintelligence.org/fomc-dashboard/index.html): A pilot dashboard comparing how FOMC governors talk about the economy — topic, narrative type, stance, and time-orientation across four speeches, with a drill-down inspector for individual expressions.
- [Chokepoint News-Risk Dashboard](https://crowintelligence.org/chokepoints/): Topic-modelled daily geopolitical risk across the world's maritime chokepoints — Suez, the Red Sea, Hormuz, Panama, the Bosphorus and the Danish straits — tracking how risk is framed in the news as it unfolds.
- [Corruption Press Networks — The K-Monitor Archive](https://crowintelligence.org/kmdb/): A data-driven analysis of K-Monitor's Hungarian corruption news archive: the most frequent and characteristic words, vocabulary richness, the co-occurrence network of the actors — people and institutions — and the types of corruption, broken down by parliamentary cycle. (In Hungarian.)
- [Felsőzsolca — A Town in Data](https://crowintelligence.org/felsozsolca/): A data portrait of a Hungarian town: warming stripes of annual mean temperature, housing and population, incomes and local economy, what changed between two censuses, and year-by-year Landsat imagery of the town and its industrial zone. (In Hungarian.) (Work in progress.)
- [Semantic Shifts in Presidential Rhetoric](https://crowintelligence.org/analysis/index.html): Tracking how the meaning of political concepts like freedom, democracy, and war evolves across 250 years of American presidential discourse. Built with our open-source chronowords package — PPMI embeddings, Procrustes alignment, and NMF topic modeling.
- [Semantic Explorer — Interactive Co-occurrence Networks](https://crowintelligence.org/semantic_explorer/app/index.html): Explore the semantic structure of classic texts through interactive ego networks. Type a seed word, click to expand, and trace how meaning propagates through Tolstoy, Plato, Adam Smith, Darwin, and more. Built with our open-source kenon package.
- [The Tractatus as a Flat Spiral](https://crowintelligence.org/aporia/wittgenstein/viz/handwriting.html): The complete text of Wittgenstein's Tractatus walked out as a flat, right-angle spiral — one continuous path through its numbered propositions, set in handwriting and coloured by topical cluster. Companion visualisation to the Aporia essay.
- [The Tractatus as a Helix](https://crowintelligence.org/aporia/wittgenstein/viz/handwriting_3d.html): The same spiral lifted into three dimensions: each proposition set a constant step above the last, so the Tractatus climbs as it turns. Rendered in handwriting and navigable sentence by sentence.
- [The Punctuation Spiral](https://crowintelligence.org/aporia/wittgenstein/punctuation_spiral.html): The Tractatus stripped to its punctuation alone — every mark in sequence, spiralled outward. What remains when the words are removed: the rhythm and breath of the argument.
- [Hungarian Pop Lyrics — A Diachronic Analysis](https://crowintelligence.org/magyar-dalszovegek/#temak): Six decades of Hungarian popular-music lyrics: distinctive words, shifting themes, semantic drift and genre from the 1960s to today. (In Hungarian.)

## Open-source Python packages

Tools built for the lab's own research and released under the MIT licence. Together they form a small stack for corpus analysis: semantic change over time, co-occurrence networks, keyness, readability, and visualisation.

- [chronowords](https://pypi.org/project/chronowords/): Detect semantic shifts over time in text corpora. Memory-efficient PPMI-based word embeddings via Count-Min Sketch, NMF topic modeling, and Procrustes alignment for tracking how word meanings evolve across time periods. (v0.2.0, Python 3.10+; source: https://github.com/crow-intelligence/chronowords; docs: https://chronowords.readthedocs.io/en/latest/)
- [kenon](https://pypi.org/project/kenon/): Construct semantic and co-occurrence networks from text using corpus-internal statistics. Lightweight graph construction through spaCy tokenization, skip-gram windows, and network backbone extraction — no neural models or external training data required. (v0.1.0, Python 3.11+; source: https://github.com/crow-intelligence/kenon; docs: https://kenon.readthedocs.io/en/latest/)
- [keyflux](https://pypi.org/project/keyflux/): Corpus keyness done properly: keywords and lockwords from a focus-versus-reference comparison (log-likelihood significance, log-ratio effect size), rank-turbulence divergence between the ranked lists, and allotaxonograph plots — a transparent, pip-installable pipeline rendered in matplotlib. (v0.1.2, Python 3.11+; source: https://github.com/crow-intelligence/keyflux; docs: https://keyflux.readthedocs.io/)
- [lexograph](https://pypi.org/project/lexograph/): Spatialize linear text into pictures you can read: punctuation spirals, 2-D/3-D sentence walks, recurrence dotplots, and concordance plots through one segment → layout → encode → render pipeline. Pure Python, matplotlib figures — the visualisation member of the corpus-lx family. (v0.1.0, Python 3.11+; source: https://github.com/crow-intelligence/lexograph; docs: https://lexograph.readthedocs.io/)
- [saphes](https://pypi.org/project/saphes/): Readability (LIX) and lexical diversity (TTR/MATTR) with the parameters other implementations hardcode. The LIX long-word threshold is fixed at 6 everywhere — a Swedish value that saturates on agglutinative or heavily inflected languages — so saphes exposes it, along with every count behind the score. Dependency-free: plain Python and the standard library. (v0.1.0, Python 3.11+; source: https://github.com/crow-intelligence/saphes; docs: https://saphes.readthedocs.io/)
- [corvus](https://github.com/crow-intelligence/corvus): A cookiecutter template for data science and text analysis projects. Pre-configured scaffold with uv, ruff, DVC, MLflow, Sphinx docs, and structured directories — eliminate manual setup and start analysing. (v1.0.0, Python 3.10+; source: https://github.com/crow-intelligence/corvus)

## Services

Commissioned work, taken on in limited volume alongside the lab's own research.

- [AI & NLP Due Diligence](https://crowintelligence.org/services/ai-due-diligence.html): Independent technical due diligence on AI and NLP startups for early-stage venture funds — technology, data, IP, team, and EU AI Act readiness. (For Venture Funds.)
- [The AI Adoption Audit](https://crowintelligence.org/services/ai-adoption-audit.html): A fixed-scope diagnostic that finds why an internal AI tool has stalled — across the data, retrieval, and cognitive layers — and sets out what to change. (For Enterprise & Scale-ups.)
- [Commissioned Analysis](https://crowintelligence.org/services/commissioned-analysis.html): Data analysis, visualisation, and visual storytelling — from a single interactive chart to a scrollable data essay — built on our own open-source tools. (Visualisation & Story.)

## About the lab

- [About](https://crowintelligence.org/about.html): what the lab is for, and full biographies of both founders.
- [Portfolio](https://crowintelligence.org/portfolio.html): the work in one place — essays, projects and packages.
- [Projects](https://crowintelligence.org/projects.html): the complete list of interactive research projects.
- [Packages](https://crowintelligence.org/packages.html): the open-source packages, with installation and documentation links.
- [Services](https://crowintelligence.org/services.html): how commissioned and advisory work is scoped.
- [Contact](https://crowintelligence.org/contact.html): enquiry form, email, and a link to book a call.

## Optional

- [Mind the Minds](https://mindtheminds.substack.com/): the lab's newsletter, on Substack. Off-site.
- [Blog](https://blog.crowintelligence.org/): older posts, hosted separately on a subdomain. Off-site.
- [Portfólió](https://crowintelligence.org/hu/portfolio.html): the portfolio page in Hungarian.
- [Privacy policy](https://crowintelligence.org/privacy.html): what data the site collects, and the cookie and analytics policy.
- [Terms](https://crowintelligence.org/terms.html): terms of use, operator identity, and licensing of the site's content.