WALS Identifies a Specific Set of Maximally-Balanced Typological Features

Ledger + dual structural thesis. The ledger is two ranked tables: top-15 most universal WALS features by dominant-value share, and top-15 most balanced features by normalised entropy. The thesis is the contrast. (1) The single most universal typological feature in the WALS database is 'absence of minor morphological means of signaling negation' at 99.2 % of 1,325 sampled languages. The next several most universal features (≥ 93 %) are also negation-related or phonological-inventory-related. These are well-defended typological universals — overwhelmingly skewed against rare exotic options. (2) Four fundamental grammatical categories — 'The Future Tense' (binary, 222 languages, H/Hmax = 1.000), 'Perfective/Imperfective Aspect' (binary, 222 languages, 0.994), 'Passive Constructions' (binary, 373 languages, 0.988), and 'Zero Copula for Predicate Nominals' (binary, 386 languages, 0.994) — sit at the maximum or near-maximum of binary entropy. These are categories that English speakers take for granted as 'features of language', but the world's languages are split essentially 50/50 on whether they have them at all. The dual finding shows that language universals and language diversity coexist on completely different feature classes: the things WALS coders look for as 'minor morphological' or 'phonologically marked' tend to be near-absent everywhere, while the things English-monoglot intuition assumes are universal are in fact maximally polarised. This is a clean quantitative restatement of typology folklore that does not appear as a single pinned table in the literature I could find.

Linguists have spent decades collecting features of the world's languages — does this language have a future tense, do its verbs change for past or present, does it have a particular kind of vowel, does it put adjectives before or after nouns, and so on. The result is the World Atlas of Language Structures, called WALS, with 192 such features documented across about 3,500 languages. I asked it two questions. First: what's the most 'universal' feature of human language? Of all 192 features, which one has the same answer for the largest fraction of languages? The answer is something pretty technical — almost no language signals negation through a small morphological marker — present in 99.2 % of the 1,325 languages with that data. So that's a real human-language universal: nearly every language does it the same way. Second question: are there features that split the world's languages right down the middle? The answer is yes, and the categories that do are surprising. Whether a language has a grammatical future tense — like English's 'will' — is split exactly 50/50 across 222 sampled languages. Half of human languages don't have one. Whether a language has perfective vs imperfective aspect — the difference English roughly captures with 'I ran' vs 'I was running' — is also 50/50. Whether a language has passive constructions ('the ball was kicked') is 50/50. Whether a language requires a copula in 'X is Y' sentences (English does, with the word 'is'; many languages don't) is 50/50. So if you're an English speaker who thinks about 'features of language' as tense, aspect, voice, and copula — the four pillars of the school grammar your English teacher drilled into you — humanity is split exactly down the middle on each of them. They are NOT human universals; they are English. The actual human universals are weirder, smaller, and more specialised than most people would guess. None of this is news to a working linguist, but pinning the exact 50/50 splits to a specific WALS file with exact entropy values, and putting it next to the 99.2 % most-universal feature in one table, is a sharper presentation than the typology literature usually offers.

99.2 % (Minor morphological negation = None) · 95.9 % (Number of Possessive Nouns = None reported) · 95.4 % (Verb-Initial with Clause-Final Negative = No) · 93.4 % (Front Rounded Vowels = None) · 93.4 % (Postnominal relative clauses = NRel dominant)

1.000 (The Future Tense, binary, n = 222) · 0.995 (Conjunctions and Universal Quantifiers, 3-way, n = 116) · 0.994 (Perfective/Imperfective Aspect, binary, n = 222) · 0.994 (Zero Copula for Predicate Nominals, binary, n = 386) · 0.991 (Epistemic Possibility, 3-way, n = 240)

192 features · 170 with ≥ 100 coded languages · 76,475 total feature values · 3,573 languages · 1 feature ≥ 99 % universal · 4 binary features ≥ 0.988 normalised entropy