Linguistic Universals¶
Core Idea¶
Linguistic Universals are the systematic empirical claim that all — or nearly all — human languages share observable structural patterns that suggest underlying principles transcending cultural or genetic specificity. The foundational construct decomposes into four inseparable components:
(1) The universal claim: the cross-language regularity proposed as universal — a structural or organizational property hypothesized to hold across the world's ~7,000 languages (e.g., every language has pronouns; every language distinguishes something like nouns from verbs; most languages have SOV or SVO word order; if a language has dominant VSO order, it has prepositions).[1]
(2) The typological evidence: the cross-language evidence — empirical data from a sample of typologically diverse, phylogenetically independent languages documenting the claimed regularity. Greenberg's 1963 foundational work[2] surveyed 30 languages across language families; modern typological databases (World Atlas of Language Structures, Universals Archive) survey hundreds. The sample must be broad and unbiased to resist sampling artifacts.
(3) The explanatory account: the explanatory account — the theoretical machinery proposed to explain why a universal obtains. Competing accounts divide into: (a) innatist explanations (Chomsky 1965, 1981[3] — universals reflect innate Universal Grammar, a language-specific cognitive module); (b) functionalist explanations (Hawkins 1994, Bybee 2010[4] — universals emerge from processing constraints, communicative pressures, or frequency effects in language use); © historical accounts (universals reflect shared inheritance, areal diffusion, or contact); (d) skeptical accounts (universals are statistical artifacts of sampling bias or analytical scheme, not genuine patterns).[5]
(4) The contested-status modulation: the absolute-vs-statistical-vs-implicational variant — the form the universal claim takes. Early Grebergian typology sought absolute universals (all languages have X); modern typology accepts statistical universals (most languages have X with frequency F); implicational universals (if language has X, then language has Y) formalize dependencies. The claim's evidential and explanatory standing shifts with its formulation.
How would you explain it like I'm…
Same things in all languages
Shared language patterns
Cross-language structural patterns
Structural Signature¶
Six italicized role-phrases identify the functional signature of linguistic universals:
- The universal claim — a proposed cross-language structural regularity, formulated as absolute, statistical, or implicational
- The cross-language evidence — empirical documentation from typologically diverse samples; must resist sampling bias and validate breadth
- The explanatory account — the theoretical framework offered (innate UG, processing/cognitive constraints, communicative function, contact/inheritance, artifact)
- The absolute-vs-statistical-vs-implicational variant — the form of the claim; early universals were absolute; modern universals are mostly statistical; implicational forms capture conditional dependencies
- The surface-vs-deep level — surface universals (word order patterns, phonological inventories) vs. deep universals (recursion, hierarchical structure, semantic composition); deeper universals are more resistant to counterexamples but harder to verify
- The contested-status modulation — the adversarial response: do exceptions overturn the universal, or refine it? How do counterexamples affect the claim's status?
A claim counts as a linguistic universal when: a cross-language sample documents a measurable regularity across independent languages; the claim is formulated precisely (absolute, statistical, or implicational); plausible explanations account for the pattern; and the claim survives scrutiny from competing explanatory frameworks and counterexample challenges.
What It Is Not¶
-
Not all linguistic regularity or cross-language similarity — universals require systematic empirical validation, broad typological sampling, and explicit claim-stating. Anecdotal cross-language similarities (both languages have a word for "mother") do not constitute universals without sampling discipline.
-
Not Universal Grammar specifically — Universal Grammar (Chomsky 1965, 1981) is one explanatory account among several. Universals can be defended without committing to UG. The prime describes the patterns; it is agnostic about the explanation.
-
Not just typology — typology is the comparative cross-linguistic method that discovers universals. Universals are the specific patterns identified; typology is the investigative framework. Related but not identical.
-
Not surface-uniformity claims — universals can be deep structural principles (recursion, hierarchical constituency) that manifest in varied surface forms. Surface diversity (SVO vs. SOV) can instantiate deep universals (hierarchical argument structure).
-
Not just statistical-tendency claims — universals require more than mere frequency. A feature present in 60% of languages might be coincidental cross-cultural practice rather than a principled universal. Universals imply underlying structural or functional explanation.
-
Not claims of historical coincidence — some language similarities reflect shared inheritance or contact (genetic relatedness, borrowing, areal diffusion). Universals claim principled regularity transcending individual histories.
Broad Use¶
-
Linguistic typology (core domain): Greenberg 1963 initiated systematic typology by cataloguing universals across 30 languages.[1] Comrie 1989 and Croft 2003 systematized typological methods and expanded universal inventory. Dryer and the World Atlas of Language Structures (Haspelmath et al.) document hundreds of structural variables and test universals cross-linguistically. Modern typology is the empirical engine for universal discovery.
-
Generative linguistics: Chomsky 1965 Aspects of the Theory of Syntax and 1981 Lectures on Government and Binding proposed Universal Grammar as an innatist explanation for linguistic structure.[6] Principles-and-Parameters approach formalized universals as core principles (structure-dependence, structure-preservation) with parameterized variation (head-initial vs. head-final). The framework explains universals as consequences of innate cognitive architecture.
-
Psycholinguistics and processing: Hawkins 1994 Performance Theory of Order and Constituency proposes that word-order universals emerge from online-comprehension efficiency.[7] Processing constraints (parsing complexity, memory load, articulatory ease) shape language structure cross-linguistically. Universals reflect general cognition plus communicative pressure, not language-specific modules.
-
Usage-based and functional linguistics: Bybee 2010 Language Usage and Cognition argues universals emerge from frequency effects, analogy, and semantic/phonological reduction in everyday language use.[8] No need for innate UG; usage patterns bootstraps structure. Modern cognitive linguistics (Lakoff, Langacker, Croft) explains universals through embodied cognition and functional motivation.
-
Evolutionary linguistics: Universals constrain hypotheses about language origins. If recursion is universal, evolutionary accounts must explain how recursion emerged. If only some languages have recursion (challenging the universality), evolutionary narratives change. Universal claims have deep implications for human cognitive evolution.
-
Language acquisition: Universals constrain the hypothesis space children explore during acquisition. Proposed universals (e.g., strong preference for consonant-initial syllables) guide developmental sequences. Universal biases reduce the learning problem (poverty of stimulus argument).
-
Cognitive linguistics and embodiment: Lakoff & Johnson 1980 and subsequent work propose that universals emerge from embodied experience: spatial orientations (UP-DOWN, FRONT-BACK) are universal because human bodies are spatially oriented; containment metaphors are universal because humans interact with containers.[9] Universals reflect embodied cognition, not innate linguistic modules.
-
Endangered-language documentation: Universals serve as discovery heuristics. When documenting an underdocumented language, fieldworkers ask: does this language have pronouns (expected universal)? How is negation marked (expected universal)? Deviations from expectations trigger investigation, illuminating language-specific properties.
-
Machine translation and multilingual NLP: Multilingual models (XLM, mBERT, XLM-R[10]) and cross-lingual transfer learning (Devlin et al. 2019 BERT pretraining[11]) exploit universals. When a universal holds across a model's training languages, transfer to an unseen language is easier. Universals enable zero-shot or low-resource cross-lingual systems.
-
Computational linguistics and linguistic structure discovery: Universal properties (part-of-speech distinctions, hierarchical constituency, recursive embedding) are built into model architectures. The assumption that universals hold accelerates learning; where universals fail or are language-specific, models struggle.
Clarity¶
Distinguishes culture-specific or language-specific quirks from deep-seated structural invariants across human linguistic systems. Without the universals/variation distinction, every feature of every language appears equally exotic, and cross-linguistic comparison lacks leverage. With the distinction, analysts identify which features transfer reliably across contexts (universals and near-universals) and which require per-language handling (language-specific properties). This clarity is useful in both basic science ("What kind of system is human language?") and applied design ("Which multilingual transfer assumptions are safe? Which fail catastrophically?").
Manages Complexity¶
Reduces the apparent diversity of 7,000 languages by identifying structural invariants and a constrained set of principled variation dimensions. Instead of modeling every language as sui generis, typologists characterize languages by their position in a parameter space defined by universal dimensions (e.g., head-directionality, null-subject licensing, tone presence/absence). This reduction substantially simplifies learning (language acquisition), teaching (language pedagogy), and engineering (multilingual NLP system design). The reduction is particularly useful in applied contexts where cross-language tooling must decide which features to assume universally and which to parameterize per-language.
Abstract Reasoning¶
Trains analysts to ask, of any complex diverse domain, what is the universal structural inventory, and what varies? The reasoning pattern generalizes beyond linguistics: cultural universals (all human societies have X), cognitive universals (all humans exhibit X), biological universals (all cells have X), and design universals (all GUI systems have navigation, selection, feedback). The discipline is to distinguish genuine structural invariants grounded in explanation from artifacts of limited sampling, conceptual scheme, or dominant-culture bias.
Knowledge Transfer¶
| Domain | Candidate universal | Evidence/Status | Notes |
|---|---|---|---|
| Morphology | Noun/verb distinction | Quasi-universal | Nearly present; exact boundaries vary |
| Syntax | All languages have clausal structure | Quasi-universal | Structure of clauses varies; all languages can embed clauses |
| Syntax | Subject precedes object (SOV, SVO, VSO dominant) | Statistical universal | Greenberg's universal #3; OSV, OVS, VOS rare |
| Semantics | Basic color terms follow Berlin-Kay hierarchy | Disputed | Cross-cultural color categorization; evolutionary/functional basis |
| Pragmatics | Address-form distinctions (polite/formal vs. familiar) | Quasi-universal | Realization varies; principle of social-status marking recurs |
| Prosody | Declarative vs. interrogative contour distinction | Quasi-universal | Specifics vary by language; prosodic role recurs |
| Phonology | Stops and nasals near-universal | Near-universal | Cross-linguistic consonant-inventory regularities |
| Semantics | Recursive embedding (e.g., in relative clauses) | Disputed (Everett 2005[^everett-pirahã]) | Pirahã and others challenged recursion universality |
| Universals | Language-internal variation (dialects, registers) | Quasi-universal | All large languages show sociolinguistic variation |
The table shows spread: some entries are well-attested language universals (basic lexical categories, clause structure, negation marking); others are contested (recursion, color categorization). Analysts must track both attested universals and challenges to universality claims. The Pirahã case (Everett 2005) exemplifies how counterexamples refine universal claims: recursion was claimed absolute; now the claim is narrower or qualified.
Examples¶
Formal/Abstract Example¶
Formal: Greenberg's 1963 "Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements" surveyed 30 typologically diverse languages and formulated 45 universals. Greenberg's Universal 3: "Languages with dominant VSO order are always prepositional" (verb-initial languages consistently use prepositions, not postpositions). Greenberg's Universal 4: "With overwhelmingly greater than chance frequency, languages with normal SOV order are postpositional" (verb-final languages show statistical bias toward postpositions). The method — broad cross-linguistic sample, explicit claim-stating, quantitative documentation — launched modern linguistic typology. Subsequent work (Dryer; World Atlas of Language Structures; typological databases) has refined, revised, and added universals. Some claims strengthen to near-absolute status; others demote to statistical tendencies with documented exceptions.[12]
Mapped back: This exemplar demonstrates the universal claim (VSO implies prepositions), the cross-language evidence (30-language sample), the absolute-vs-statistical variant (some stated as absolute, others as strong statistical biases), and the explanatory account (Greenberg proposed functional grounding in speech-production efficiency).
Applied/Industry Example¶
Non-formal, structurally faithful: An international software-engineering team designs a multilingual NLP pipeline for sentiment analysis. The team audits which linguistic properties generalize across their 24 target languages and which require language-specific handling:
-
Near-universals assumed globally: part-of-speech categories (noun, verb, adjective, preposition); hierarchical syntactic embedding; semantic compositionality (meaning of a phrase is a function of the meanings of its words). These universals are built into the model architecture (transformer-based, hierarchical attention).
-
Statistical universals tested individually: word-order patterns (SVO languages use different parsing strategies than SOV); null-subject licensing (pro-drop languages like Spanish license null subjects; non-pro-drop languages like English do not). The team's multilingual model (mBERT, XLM-R) has learned these patterns from training data; transfer works when universals hold, fails when language-specific properties dominate.
-
Language-specific properties handled per-language: morphological complexity (Swahili has complex agglutination; English has minimal inflection); tone (Mandarin is tonal; English is not); honorific systems (Japanese has elaborate politeness-marking; English minimal). The team either builds language-specific modules or fine-tunes shared models with language-specific data.
The result: a sentiment-analysis system that leverages universals for zero-shot cross-lingual transfer, tests universals empirically (transfer success validates the universal; transfer failure challenges it), and gracefully degrades to language-specific handling where universals fail or are inapplicable.
Mapped back: This exemplar shows the universal claim (part-of-speech distinctions, compositionality transfer cross-linguistically), the cross-language evidence (24-language test suite demonstrates empirically which universals hold), the explanatory account (computational: universals are shared architectural principles), the contested-status modulation (some universals are validated by transfer success; others are challenged when transfer fails), and the absolute-vs-statistical variant (compositionality is assumed absolute; word-order effects are statistical).
Structural Tensions¶
T1 — Innate Universal Grammar vs. Functional Explanations. Chomsky's UG (1965, 1981) posits that linguistic universals reflect innate biological endowment — a language-specific cognitive module.[13] Competing functionalist accounts (Hawkins 1994 processing efficiency, Bybee 2010 usage-based learning, Croft 2003 cognitive linguistics) explain universals as emergent from general cognition, communicative pressures, frequency effects, and embodied experience. The tension is foundational: Are universals biologically determined or functionally emergent? Implications: if innate, language is modular and partially immune to experience; if functional, language is grounded in general cognition and shaped by usage. Evidence cuts both ways; neither position has achieved consensus.
T2 — Absolute vs. Statistical Universals. Early Grebergian typology sought absolute universals (all languages have X). Accumulating evidence revealed most universals are statistical: word order tends toward SVO/SOV but not exceptionlessly; most languages have stops but not all; most languages have color-term hierarchies but not universally. Modern typology accepts statistical formulations ("X tends to co-occur with Y in 80% of cases") and implicational forms ("If X, then usually Y"). The tension is whether apparent exceptions overturn a universal claim or merely refine it to statistical status. Implications: absolute universals are stronger claims (explain more, predict more) but rarer; statistical universals are weaker but more defensible empirically.
T3 — Counterexamples and Universal Status. Everett 2005 challenged the universality of recursion in Pirahã; subsequent work questioned other claimed universals (color-term hierarchy, recursion in complex structures). The tension is how to treat counterexamples: Does a single counterexample overturn a universal claim? Or do exceptions refine the claim (recursion is near-universal with documented exceptions)? Do counterexamples expose sampling bias (Pirahã is an exceptional language), or do they reveal the universal was never truly universal? Implications: how we treat exceptions determines whether universals are falsifiable (Popperian) or refined into near-tautologies; careful formulation distinguishes robust patterns from spurious generalizations.
T4 — Surface vs. Deep Universals. Surface universals (word order patterns, phonological inventories, morphological complexity) are easy to measure but vary widely across languages. Deep universals (hierarchical recursive structure, semantic composition, principles of recursion) are harder to identify empirically but more resistant to counterexamples (nearly all languages can embed clauses recursively even if surface order differs). The tension is between observability and robustness: surface universals are visible but fragile; deep universals are more fundamental but harder to verify without explicit theory. Implications: descriptive typology focuses on surface universals; theoretical linguistics probes deeper structures. Both are needed; each constrains the other.
T5 — WEIRD-Language Sampling Bias. Typological samples have historically been biased toward well-studied, documented languages (European, Asian, a handful of African and Pacific languages). Bender 2011 ("Every language matters") and subsequent critical work highlight sampling bias: statistical universals may reflect the sample composition rather than genuine cross-linguistic patterns. Modern typology expands sampling to endangered languages, understudied language families, and polysynthetic languages. The tension is whether observed universals are genuine or artifacts of sampling: languages excluded from early samples might violate the universal. Implications: universal claims must be validated on representative, unbiased samples; WEIRD-language bias is a serious threat to universals claims; expansion to understudied languages is essential for validation.
T6 — Empirical NLP Validation vs. Theoretical Typology. Modern multilingual NLP (cross-lingual transfer learning, multilingual BERT, XLM-R) provides an empirical test of universals: when a linguistic universal holds across languages, transfer learning succeeds (model trained on English transfers to Japanese); when universals fail or are language-specific, transfer breaks down. This provides new empirical traction on universals claims. The tension is between computational-empirical validation and traditional typological argumentation: Do NLP transfer-learning results validate or challenge universals? Can model success predict linguistic universality? Implications: NLP provides large-scale empirical tests of universals; some universals (word-order correlations) are validated empirically; others (recursion universality, color-term hierarchies) remain contested even with NLP evidence. The two methodologies (traditional typology vs. computational testing) are converging but sometimes in tension.
Structural–Framed Character¶
Linguistic Universals is a hybrid on the structural–framed spectrum. Part of it is a bare pattern that means the same thing in any field — a regularity claimed to hold across all members of a class, possibly in implicational form (if a system has feature X it also has Y); part of it is a frame, a vocabulary of grammar, language, and cognition, inherited from linguistics.
The structural skeleton transfers cleanly: the logic of a cross-case universal, whether absolute, statistical, or implicational, applies just as well to recurring features of biological organisms, legal systems, or musical traditions as it does to language. But the prime is tied to its home discipline in substance. Its canonical examples and vocabulary — phonemes, word order, grammatical categories, the debate over innate principles across the world's roughly seven thousand languages — are specifically linguistic, and it carries assumptions about human cognition and what counts as a language. Its origin is an empirical research program in linguistics, not a purely formal relation, and applying it elsewhere means importing that comparative-grammar framing. With a portable abstract core but a substantial inherited frame, it sits on the framed side of the middle of the spectrum.
Substrate Independence¶
Linguistic Universals is a moderately substrate-independent prime — composite 3 / 5 on the substrate-independence scale. These are cross-language patterns proposed as fundamental to human language cognition, and although the claim is genuinely empirical — testing against some seven thousand languages — the concept and its evidence remain firmly grounded in linguistics and cognitive science. Its signature is linguistically flavored, worked examples are absent, and any reach into non-linguistic domains like logic, music, or formal systems is metaphorical rather than structural. That confinement to one substrate, with only occasional metaphorical extension, is what holds it to the middle of the scale.
- Composite substrate independence — 3 / 5
- Domain breadth — 3 / 5
- Structural abstraction — 3 / 5
- Transfer evidence — 2 / 5
Relationships to Other Primes¶
Parents (1) — more general patterns this builds on
-
Linguistic Universals is a decomposition of Invariance
Linguistic universals is the structurally-particularized form invariance takes in the cross-linguistic case: the preserved feature is the named structural property (e.g., "every language distinguishes nouns from verbs"), the transformation group is variation across the ~7,000 languages, and the empirical evidence is the typological sample. It satisfies invariance's joint commitment — what is preserved under which operations — particularized to the case where the operations are inter-linguistic comparison and the invariants are the universals themselves.
Path to root: Linguistic Universals → Invariance
Neighborhood in Abstraction Space¶
Linguistic Universals sits among the more crowded primes in the catalog (6th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.
Family — Language, Symbol & Cultural Form (32 primes)
Nearest neighbors
- Iconicity — 0.86
- Paradigmatic vs. Syntagmatic Relations — 0.85
- Simile — 0.84
- Deixis — 0.84
- Emergent Formalization (Language) — 0.83
Computed from structural-signature embeddings · 2026-05-29
Not to Be Confused With¶
Linguistic Universals differs fundamentally from Paradigmatic vs. Syntagmatic Relations, despite both operating on structural axes within language. Paradigmatic and syntagmatic relations describe internal organizational mechanisms within a single language's structure. Paradigmatic relations organize elements into contrast sets (nouns, verbs, adjectives in the paradigm slot; they can substitute for one another in a syntactic position). Syntagmatic relations organize elements in sequence (subject precedes verb precedes object; words combine linearly). These axes are inherent to how any language organizes meaning at a given moment. Linguistic Universals, by contrast, describe cross-language regularities—patterns that recur across the 7,000 human languages, suggesting principled constraints transcending individual language structure. A language's paradigmatic organization (what contrast sets it maintains) and syntagmatic ordering (which sequence patterns it allows) are language-specific implementation details. The universal claim points to regularities above those details: which contrast sets recur (noun/verb distinctions appear universally), which ordering patterns recur (SOV and SVO dominate globally), and why. Paradigmatic-syntagmatic relations explain how language organizes internally; linguistic universals explain what patterns appear consistently across languages. A linguist analyzing English paradigmatic structure (pronouns I, me, my, mine form a paradigm; they occupy the same syntactic slot) is doing structural analysis. A typologist documenting that all languages mark agent-patient distinction in some form is describing a universal. Both are valid linguistic inquiry, but they operate at different scopes: internal organization versus cross-language constraint.
Linguistic Universals also differs from Markedness, which describes asymmetry between paired elements within a language. Marked forms carry explicit features (feminine -e in French; emphatic stress patterns); unmarked forms lack those features and are cognitively simpler, more frequent, and acquired earlier. Markedness is a relationship between forms in a single language. One element in a pair is marked (bears special features), the other unmarked (the default or base form). Linguistic Universals, by contrast, describe regularities across languages: which features appear cross-linguistically, which orderings dominate, which structural dependencies recur. Markedness and universals intersect—marked forms may show cross-linguistic patterns, and universals may predict which form in a marked/unmarked pair appears in all languages (agents precede patients across languages; the agent slot is pragmatically unmarked, the patient marked). But the primes are distinct: markedness asks "which form in this pair is structurally prominent, frequent, and assumed as default?"; universals ask "which features or structures appear across all languages or most languages?" A gender system (French masculine/feminine) exemplifies markedness within French: one form is marked, one unmarked. The universal claim that most languages mark gender in some form is a linguistic universal. Markedness is the internal mechanism; universals describe cross-language regularities in how those mechanisms are implemented.
Linguistic Universals is wholly distinct from Universality in Critical Phenomena, which transfers the concept from physics to linguistics metaphorically but names a different mathematical phenomenon. Universality in critical phenomena describes how systems near phase transitions (boiling points, magnetic transitions, network collapses) exhibit identical scaling behavior despite arising from completely different substrates (water, iron, social networks). The mathematical behavior—power laws, critical exponents—is universal across these disparate systems. Linguistic universals name empirical cross-linguistic patterns: all languages have nouns; most have subject-object distinction; word-order preferences correlate with morphology. These are regularities in human language structure, not scaling behaviors near phase transitions. While one could metaphorically apply critical-phenomena language to language-change thresholds ("When does a dialectal shift trigger language split? Do critical-phenomena power laws apply?"), the core prime—linguistic universals—is an empirical typological claim about which features recur cross-linguistically. Critical-phenomena universality is a mathematical scaling property; linguistic universals are structural and organizational regularities. The former is substrate-independent in a precise mathematical sense; the latter is domain-specific to human language cognition and its uses.
Solution Archetypes¶
Solution archetypes in the catalog that build on this prime — directly (this prime is a source ingredient) or as a related prime.
Built directly on this prime (1)
Also a related prime in 1 archetype
References¶
[1] Greenberg, J. H. (1963). Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements. In J. H. Greenberg (Ed.), Universals of Language (pp. 73–113). MIT Press. Greenberg Universals of Language foundational typological method implicational universals. ↩
[2] Greenberg, J. H. (1963). Universals of Language. MIT Press. Greenberg 30-language sample typological method universal discovery empirical validation. ↩
[3] Chomsky, N. (1965). Aspects of the Theory of Syntax. MIT Press. and Chomsky, N. (1981). Lectures on Government and Binding. Foris Publications. Chomsky Aspects Lectures Government Binding Universal Grammar innatist Principles Parameters. ↩
[4] Hawkins, J. A. (1994). A Performance Theory of Order and Constituency. Cambridge University Press. and Bybee, J. L. (2010). Language, Usage and Cognition. Cambridge University Press. Hawkins Bybee processing constraints frequency usage-based functional explanation universals. ↩
[5] Multiple explanatory frameworks for linguistic universals: innatism (Chomsky UG), functionalism (Hawkins, Bybee), historical inheritance, contact/areal diffusion, and skeptical artifact accounts. linguistic universals explanatory accounts competing theoretical frameworks. ↩
[6] Chomsky, N. (1965). Aspects of the Theory of Syntax. MIT Press. and Chomsky, N. (1981). Lectures on Government and Binding. Foris Publications. Chomsky Universal Grammar innatist explanation linguistic structure Principles Parameters. ↩
[7] Hawkins, J. A. (1994). A Performance Theory of Order and Constituency. Cambridge University Press. Hawkins Performance Theory word-order universals processing efficiency comprehension. ↩
[8] Bybee, J. L. (2010). Language, Usage and Cognition. Cambridge University Press. Bybee Language Usage Cognition frequency effects analogy semantic-phonological reduction bootstrapping. ↩
[9] Lakoff, G., & Johnson, M. (1980). Metaphors We Live By. University of Chicago Press. Lakoff-Johnson Metaphors We Live By embodied cognition spatial metaphors universals. ↩
[10] Conneau, A., Khandelwal, K., Goyal, N., et al. (2020). Unsupervised Cross-lingual Representation Learning at Scale. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (pp. 8440–8451). Conneau XLM-R multilingual model cross-lingual transfer learning universals. ↩
[11] Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of NAACL-HLT 2019 (pp. 4171–4186). Devlin BERT pre-training deep bidirectional transformers multilingual representation. ↩
[12] Greenberg, J. H. (1963). Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements. In J. H. Greenberg (Ed.), Universals of Language (pp. 73–113). MIT Press. Greenberg Universal 3 4 word-order universals prepositional postpositional implicational. ↩
[13] Chomsky, N. (1965). Aspects of the Theory of Syntax. MIT Press. and Chomsky, N. (1981). Lectures on Government and Binding. Foris Publications. Chomsky Universal Grammar biological endowment language-specific cognitive module innatist. ↩
[14] Comrie, B. (1989). Language Universals and Linguistic Typology: Syntax and Morphology (2nd ed.). University of Chicago Press. and Croft, W. (2003). Typology and Universals (2nd ed.). Cambridge University Press. Comrie Croft Typology Universals systematic typological method universal catalogue.