What one row means
Grambank records language varieties, which can include closely related dialects. A point stands for a coded variety, not one speaker, country, or population.
The Shape of a Sentence
See how languages arrange words, mark meaning, and ask questions.
Try the project ↓How to use this project
Help
How to explore
Languages can express the same idea with different word orders, verb forms, pronouns, and question patterns. Grambank records these structures for 2,467 language varieties, including some closely related dialects.
Pick a language
Start with English or search by name, then read its yes, no, and missing answers for 20 selected grammar questions.
Map one answer
Choose a question and compare recorded yes shares across six broad regions without treating each point as a speaker population.
Compare two languages
Count the same answers within the selected 20 questions and keep the number of comparable answers visible.
Pair two questions
Read a four-cell count of which answers appear together, while treating the pattern as an association rather than a rule or cause.
- Language variety
- One named language or dialect entry in Grambank; it is a database record, not a count of speakers.
- Grammar feature
- One coded question about a structural part of a language, such as word order or how a question is marked.
- Language family
- A group classified as descending from a shared earlier language; this project uses Grambank's Glottolog-based family field.
- Macroarea
- One of six broad world regions used by Glottolog to organize language locations.
- Comparable answer
- A selected grammar question with a recorded yes or no for both language varieties being compared.
- Missing answer
- A question without a code for that variety; missing does not mean the feature is absent.
Grambank · 2,467 language varieties
The Shape of a Sentence
See how languages arrange words, mark meaning, and ask questions.
Languages can use very different grammar even when they express the same idea. This project follows 20 questions from Grambank. It looks at structure, not vocabulary or how many people speak a language.
01
Pick a language
Read its 20 recorded answers
Choose a language, then select any answer to carry it into the map. “Yes” means Grambank recorded the feature. It is not a score or a judgment.
Selected language variety
English
- Family
- Indo-European
- Broad region
- Eurasia
- Recorded here
- 20 of 20 answers
Try “Spanish,” “Swahili,” “Navajo,” or “Japanese.”
Nouns and pronouns
Order and position
Verbs and clauses
Questions and counting
02
Map one answer
Where is “Verb in the middle” recorded?
Each mark is one language variety at its single Grambank coordinate. The map does not show borders, speaker populations, or the full area where a language is used.
English
YesThe usual transitive clause puts the verb between its main arguments.The coral ring marks this variety’s recorded coordinate when it is inside the active region.
Recorded answers by broad region
Bars use only languages with a yes-or-no answer. Counts show coverage.
03
Compare two languages
Count the same recorded answers
This is a comparison of 20 chosen questions, not a measure of whole-language similarity. Rankings require at least 16 questions recorded for both languages.
Ranked by number of same answers, then by the number compared.
SelectedEnglish
18same of 20
Compared withWestern Frisian
04
Pair two questions
See which answers appear together
Choose a second question. The four boxes count only language varieties with both answers recorded. A pattern here does not prove a rule or explain why it exists.
First questionVerb in the middle
When verb in the middle is recorded as yes, 82% also record prepositions as yes. When the first answer is no, that share is 30%.
2,163 language varieties have both answers.Prepositions
Verb in the middle
No
Yes
No
781no + no
332no + yes
Yes
190yes + no
860yes + yes
How to read the project
A record of descriptions, not a language census
What “missing” means
A missing answer means the database does not provide a code for that variety and feature. It does not mean the feature is absent.
What the 20 questions leave out
Grambank contains 195 questions. This interface keeps 20 readable binary questions, so every comparison is a narrow view of grammar.
Why family and place matter
Shared answers can reflect history, contact, chance, or coding choices. This project shows patterns but does not assign causes.
Source snapshot and method
The static snapshot contains 2,467 language varieties, 195 Grambank questions, and 362,025 recorded values from version 1.0.3. A deterministic build selected 20 binary questions and preserved every language’s name, family, broad region, coordinate, yes/no answer, and missing answer for those questions. The browser makes no request to Grambank, Zenodo, or GitHub.
Why this exists
The question
A guided comparison of 20 clearly defined Grambank features can make structural variety visible without treating a partial database as a complete grammar, a language census, or a measure of linguistic value.
What to try
Start with English, choose one recorded grammar question, map its yes, no, and missing answers, compare the same 20 questions with another language variety, and pair two questions in a four-cell count.
Method and limits
About the data
I found Grambank v1.0.3 in the allowlisted TidyTuesday catalog's November 3, 2026 release. I opened the underlying Zenodo record and the Grambank download page before using it. The Grambank Consortium publishes the dataset under Creative Commons Attribution 4.0. The source is a structural database: it records coded grammar questions, language names, Glottolog-based family classifications, broad regions, and one coordinate per language variety. It does not measure vocabulary, speaker populations, fluency, prestige, correctness, or language quality.
The source snapshot is pinned to TidyTuesday commit `f422450f6519830f6ce0312e5294e3cf110c72ed`, which builds its three tables from the official Zenodo release. The language, parameter, and value CSV files contain 293,398, 954,917, and 9,050,664 bytes. Their SHA-256 values are `7507f95aabc6ba9e80d6556b55ea52f38f59e0fe3f0f02619d8e23570c3e3e77`, `8d5e4005f473b9b532e5a43406249529ecc830db9ec6d932b45ce5163716bd10`, and `7130ca062899e28da6dc3d27d3ef73dee5062e098eeae74f67d9f64528900e1c`. The reducer checks those exact bytes, the required identifiers, 2,467 unique language varieties, 195 unique questions, 362,025 values, six macroareas plus four entries without one, and binary values for every selected question. Repeated runs produced the same 234,161-byte browser snapshot at SHA-256 `fdac851219a064cbc72e1a792febf115193f988f44aefb758aa8168bff9691c8`. The browser makes no request to Grambank, Zenodo, TidyTuesday, or GitHub.
The interface keeps 20 binary questions chosen for readable everyday explanations and coverage across nouns, pronouns, word order, verbs, clauses, questions, and counting. Yes means the source records the feature as present; no means it records the feature as absent; missing means no value is provided. Six multistate questions and the other 169 binary questions stay outside this interface. The language comparison counts only questions with yes or no for both varieties. Its ranking requires at least 16 comparable questions and orders by the number of same answers, then by the number compared. This deliberately avoids presenting a high percentage built from only a few shared records.
The coordinate map uses the single latitude and longitude attached to each Grambank variety. It is not a territorial map and does not show where every speaker lives. Regional bars count records, not people or countries. Family labels follow the source's Glottolog-based classification, including its treatment of creoles and dialects. The two-question table is descriptive: shared answers may reflect descent, language contact, geography, sampling, documentation, coding choices, or chance. The project does not test those causes or claim a universal rule. Grambank's accompanying 2023 paper reports about 24% missing data in the full version-one database and explains that its published analyses apply additional pruning or imputation for specific questions; this project does neither. It preserves missing answers instead.
Skills used
- Data analysis
- Data engineering
- Statistical reasoning
- Systems design
- Visualization
- Information design
- Interaction design
- Visual design
- Creative coding
- Product thinking
- Editorial storytelling
- Accessibility
- Performance
What the data shows
- Grambank v1.0.3 contains 2,467 language varieties and 362,025 recorded values across 195 grammar questions. This interface keeps 20 binary questions, and 441 varieties have an answer to all 20.
- English has all 20 selected answers recorded. Haitian and Western Frisian each match 18 of those 20 answers; Emai and Tikar, from the Atlantic-Congo family, each match 17 of 20. These counts describe this selected slice, not overall similarity or common descent.
- Among 2,163 varieties with both answers, 860 of 1,050 varieties recorded with verb-medial order also record prepositions, or 81.9%. The share is 332 of 1,113, or 29.8%, where verb-medial order is recorded as absent. This is an association in the database, not a grammatical rule or causal result.
- For the decimal-number-system question, 440 of 483 coded Eurasian varieties record yes, compared with two of 84 coded Australian varieties. The comparison counts database entries and leaves 80 Eurasian and 59 Australian entries without an answer; it does not weight languages by speakers.