The Root Concordance of the Qur’an

All 1,651 triliteral roots, their dictionary words, their distribution across the text, and four measurements built on that distribution.

1. Method

Each of the 130,030 morphological segments of the Qur’an carries a root where one applies. Counting proceeds from that annotation rather than from spelling, so that every derived form of a root is gathered automatically and no unrelated word is admitted by a coincidence of letters. Where a root yields several distinct dictionary words — and many do — the lemmas are listed separately in the concordance so that a root total is never mistaken for a count of one word.

1.1 What 1,651 counts, and why published totals differ

The inventory here divides as 1,608 triliteral roots, carrying 50,156 occurrences, and 43 quadriliteral roots carrying 113. Quadriliterals are a small but real part of the vocabulary — hudhud (the hoopoe of Sūrah an-Naml), dirham, surādiq, qiṭmīr — and they are counted.

This figure will not match every published one, and that is expected rather than a discrepancy to be reconciled. Totals in circulation include about 1,650, about 1,685 and about 1,790. They differ because counting roots requires editorial decisions, and reasonable scholars decide differently:

• Are quadriliterals counted? Excluding them here would give 1,608 rather than 1,651.
• Do roots attested only in proper or foreign nouns count? Names and loanwords can be assigned a root or left unrooted.
• How are weak and geminate roots handled? Whether a hollow or doubled root is listed once or split by its variants moves the total.
• Which annotation edition? Morphological databases are revised, and a later revision may split or merge an entry.

So the number reported here is not a claim about the Arabic language. It is a precise statement of what this annotation contains, which is the only thing a count can honestly be. Every figure in this document is derived from that same inventory, so they are consistent with one another whatever another edition would give.

Sūrahs are classified Makkan or Madinan by the standard designation carried in the text’s metadata (86 Makkan, 28 Madinan), and the traditional order of revelation is used for the accumulation curve. Both are scholarly conventions rather than measurements, and section 6 says what follows from that.

2. Roots occurring exactly once

399 roots — 24.2% of the total — occur a single time in the whole Qur’an. This is the quantitative counterpart of the classical concern with gharīb al-Qur’ān, the rare vocabulary that attracted its own commentary literature. Each is given with its single location.

3. The lexical signature of Makkan and Madinan revelation

The two bodies of revelation differ in vocabulary, and the difference can be quantified. Roots are ranked by weighted log-odds with a Dirichlet prior, which measures how much more strongly a root belongs to one corpus than the other while shrinking rare roots toward no preference — a plain ratio would make every scarce root appear maximally distinctive. The Makkan corpus carries 79,642 segments, the Madinan 50,388.

Roots confined to one corpus need care. Taken at face value, 552 roots appear only in Makkan sūrahs and 200 only in Madinan ones. But a root occurring once in the whole Qur’an is exclusive to whichever corpus that single verse happens to sit in, which says nothing about a division of vocabulary — and 51% of the Makkan figure and 59% of the Madinan figure are of exactly that kind.

The substantive figures are therefore the ones that exclude single occurrences: 271 roots recur in Makkan sūrahs and never appear in Madinan ones, and 82 the reverse. Those are the numbers used throughout this document. The tables below rank roots occurring at least eight times overall, so the ranking reflects preference rather than scarcity.

3.1 Most characteristically Makkan

3.2 Most characteristically Madinan

3.3 What the two lists contain

The ranking was produced without reference to meaning: it sees only how often each root falls in one corpus rather than the other. What it returns, however, is not an arbitrary set. The two lists divide along a clear semantic line, and that is the substantive finding of this section.

The Madinan list is dominated by the vocabulary of a community with a law and a polity. It contains the roots behind qitāl (fighting), munāfiq (the hypocrite), tawba (repentance), nisāʾ (women), ḥalāl (the lawful), maʿrūf (the recognised good), ʿafw (pardon), ḥukm (judgement) and kitāb (writing, and what is prescribed). One entry is particularly telling: ʾ-L-W, the root of īlāʾ, the oath of marital abstention — an institution regulated in Madinan legislation and with little occasion to arise before it.

The Makkan list is dominated by the vocabulary of proclamation and dispute over it. It contains the roots behind rabb (Lord), ʿibāda (worship), khalq (creation), nadhīr (the warner), waḥy (revelation), qurʾān (recitation), takdhīb (denying as false), mujrim (the criminal) and siḥr (sorcery) — the last being the charge the early opponents are repeatedly said to have levelled.

That a purely frequency-based procedure recovers this division is worth stating. The Makkan and Madinan categories are traditional, and the difference between them has long been described in qualitative terms. What is shown here is that the difference is also present in the raw distribution of content vocabulary, strongly enough to be recovered by counting alone, with a magnitude attached to each root.

4. How the vocabulary accumulates

Reading the sūrahs in the traditional order of revelation, the curve below shows how many distinct roots have appeared at each step. The bars show how many roots each sūrah introduces for the first time. A steep early rise and a long flattening is what one expects of any natural corpus; what is measurable here is the shape and where the introductions cluster.

5. The frequency distribution

Plotting each root’s rank against its number of occurrences, on logarithmic axes, gives a near-straight line — the shape associated with Zipf’s law, and the shape natural language corpora generally show. The least-squares slope is -1.567 with R² = 0.9233. The commonest root occurs 2,851 times; the median root occurs 5 times.

Dotted line: least-squares fit to log₁₀(rank) against log₁₀(frequency).

6. Limitations

• Lemmatisation follows one scheme. The annotation used here assigns forms to lemmas in a particular way; another scheme would split or merge some lemmas differently, changing lemma-level figures though not root-level ones.
• Makkan and Madinan are traditional designations, not properties measured from the text. Some sūrahs are held to contain verses of the other type. The signature in section 3 therefore describes the corpora as conventionally divided.
• The order of revelation is likewise traditional and not uniformly agreed. Section 4 measures accumulation under that ordering, and a different ordering would give a different curve.
• Frequency is not significance. Nothing here should be read as inferring importance from a count.

7. Prior work and sources

This document is a measurement of the text and a resource built on published annotation. It stands on the following, and any use of it should engage them directly rather than through this summary.

7.1 Sources of the data

1. The Qurʾānic text, Uthmanī orthography, 6,236 āyāt, from Tanzil.
2. Morphological annotation — root, lemma and part of speech for each of 130,030 segments — from the Quranic Arabic Corpus (Kais Dukes, University of Leeds).
3. English quotations, where used, from Marmaduke Pickthall (1930) and Abdullah Yusuf Ali (1934), both in the public domain.

7.2 Prior quantitative work on the same question

1. Behnam Sadeghi, “The Chronology of the Qurʾān: A Stylometric Research Program”, Arabica 58:3–4 (2011), 210–299. The principal stylometric study: seven phases tested against mean verse length and the frequencies of the 28 commonest morphemes.
2. Mehdi Bazargan, foundational work on verse length as a chronological marker, on which later stylometry builds.
3. Andrew G. Bannister, Retelling the Tale: A Computerised Oral-Formulaic Analysis of the Qurʾan — finds Madinan sūrahs substantially more formulaic than Makkan.
4. Subsequent work revising the inner-Qurʾānic chronology from mean verse lengths and the Medina I counting system.
5. “Stylometry and the Refinement of a Qurʾanic Timeline”, in The Early Qurʾan in Mecca (Cambridge University Press).

7.3 Method borrowed from outside the field

1. Burt L. Monroe, Michael P. Colaresi and Kevin M. Quinn, “Fightin’ Words: Lexical Feature Selection and Evaluation for Identifying the Content of Political Conflict”, Political Analysis 16:4 (2008), 372–403. The weighted log-odds estimator with an informative Dirichlet prior, used in section 3 to rank distinguishing vocabulary without letting rare roots dominate.
2. Aaron Clauset, Cosma Rohilla Shalizi and M. E. J. Newman, “Power-Law Distributions in Empirical Data”, SIAM Review 51:4 (2009), 661–703. The reason section 5 declines to present its least-squares slope as an estimated exponent.

Citations are given so a reader can find the work. The authors of items 7.2 and 7.3 have no connection with this document and bear no responsibility for it.

8. The concordance

All 1,651 roots, ordered by frequency. The bar shows the Makkan share of each root’s occurrences. Type in the box to filter by root or transliteration.