Relative Clauses

How a relative clause works

Take the English sentence I bought a book yesterday. To use it as a modifier of book, English wraps it in a relative clause:

• the book which I bought ___ yesterday
• the book that I bought ___ yesterday
• the book I bought ___ yesterday

The blank is the gap — the object position of bought where book is interpreted. The relative clause restricts the reference of the head noun: not just any book, but the one with the property of being something I bought yesterday. Generative grammar treats the construction as wh-movement of a relative operator from the gap site to the clause edge, leaving a trace; Head-Driven Phrase Structure Grammar treats it as a gap-feature percolating up the tree; Combinatory Categorial Grammar handles it via type-raising and function composition. Whatever the formalism, two ingredients are essential: a head (the modified noun) and a gap (the relativised position).

Three relativisation strategies

Languages assemble these ingredients in three principal ways.

Strategy	Marker on the relative clause	Position of the head	Example languages
Wh-relative (with relative pronoun)	Wh-pronoun fronted to clause edge	External, post-nominal	English, German, French, Russian
Gap relative (no relative pronoun)	Bare gap, no morphological marker	External, pre- or post-nominal	Japanese, Mandarin, Korean, Turkish
Resumptive relative	Overt pronoun in the gap site	External, post-nominal	Hebrew, Irish, Welsh, Persian, Standard Arabic
Internally headed relative	Head sits inside the clause	Internal	Quechua, Navajo, Lakhota, Tibetan
Correlative	Head-clause pair connected by demonstrative	External + clause-initial	Hindi-Urdu, Sanskrit, Bambara, Hittite
Participial relative	Non-finite participle modifying the noun	External, pre-nominal	Turkish, Mongolian, Latin (in part)

The four-way split between gap, wh-pronoun, resumption, and internal-head is robust: most languages choose one as their primary strategy. English mixes wh-relatives and gap relatives (the bare-relative the book I bought is gap-only); Hebrew mixes gap and resumption depending on the relativised position.

English wh-relatives, that-relatives, and bare relatives

English has three relativisation forms in modern use.

• Wh-relative. A wh-pronoun (who, which, whose, where, when, why) fronts to the clause edge. The man who I met; the city where I lived. Mandatory in non-restrictive clauses.
• That-relative. The complementiser that introduces the clause; no wh-pronoun. The man that I met. Restrictive only.
• Bare relative. No marker at all — only a gap. The man I met. Restrictive only, and only when the gap is in object or oblique position (subject gaps require a marker: the man who/that met me, not *the man met me).

The three forms have distinct historical sources. That-relatives are inherited from Old English þæt, a generalised relative complementiser. Wh-relatives entered English through borrowing from Latin and French in late Middle English; before c. 1200 English barely used them. The relative which for animates is a 19th-century shibboleth — older English happily used which for people (Our Father which art in heaven). The bare relative is structurally older still and shows up in many varieties of spoken English.

Japanese: prenominal gap relatives

Japanese relative clauses precede the head noun and contain neither a wh-pronoun nor a complementiser. Only a gap signals relativisation.

	Romanised	Gloss
The book I bought	watashi-ga katta hon	[I-NOM bought] book
The man who came	kita hito	[came] person
The book whose author I know	chosha-o shitte-iru hon	[author-ACC know-ing] book
The day on which I left	watashi-ga shuppatsu-shita hi	[I-NOM departure-did] day

The verb in the relative clause appears in the same plain form it would take in a main clause; no participle, no relativiser. The "gap" is recovered from context plus argument structure: if a transitive verb's clause lacks an overt object, and the head noun makes sense as the object, the parser fills it in. Mandarin uses the same gap strategy but adds a marker de at the clause edge: wo mai de shu "[I bought] DE book". Korean similarly adds a relative-clause-marking suffix on the verb (-n for past, -nun for non-past).

Head-final gap relatives raise a parsing puzzle. The listener encounters the relative clause first and only later discovers the head noun — and the role the gap plays. Joan Bresnan and others note that this should be hard to parse, but Japanese listeners manage; recent work in psycholinguistics (Yamashita 1997, Aoshima et al. 2004) suggests prediction-based parsing handles the load.

Hebrew: resumptive pronouns and the gap-resumption split

Hebrew relative clauses are introduced by the relativiser she- (or, in Biblical and formal registers, asher). Some positions take a gap, others take a resumptive pronoun.

Position relativised on	Hebrew form	Strategy
Subject	ha-yeled she-ra'a oti	Gap (no pronoun)
Direct object	ha-yeled she-ra'iti (oto)	Optional resumptive
Indirect object	ha-yeled she-natati lo et ha-sefer	Obligatory resumptive (lo "to him")
Oblique	ha-yeled she-dibarti ito	Obligatory resumptive (ito "with him")
Genitive	ha-yeled she-ha-em shelo	Obligatory resumptive (shelo "his")
Object of comparison	ha-yeled she-ani gavoah mimenu	Obligatory resumptive (mimenu "from him")

The pattern follows Keenan and Comrie's Accessibility Hierarchy. Higher positions (subject, direct object) use the gap strategy; lower positions (oblique, genitive, comparison) require resumption. Irish, Welsh, and Persian show similar gradients. The cross-linguistic finding is that resumption rescues gaps that would otherwise violate constraints — long-distance gaps, gaps inside islands, gaps in deeply embedded positions. English mostly forbids resumption, but it surfaces in colloquial speech as a repair: the man that I don't know what he does, where a gap inside the indirect question would be ungrammatical.

The Accessibility Hierarchy

Edward Keenan and Bernard Comrie's 1977 paper "Noun Phrase Accessibility and Universal Grammar" surveyed fifty languages and proposed that syntactic positions are ordered for relativisability:

1. Subject
2. Direct Object
3. Indirect Object
4. Oblique (object of preposition/postposition)
5. Genitive (possessor)
6. Object of Comparison

The hierarchy predicts two implicational universals. First, if a language can relativise on a lower position, it can relativise on every higher position. Second, the same language may use different strategies at different positions — typically gap-strategy higher up, resumption lower down. Languages cluster: English relativises every position with gap-or-wh; Malagasy relativises only subjects; Tagalog requires the relativised position to be the syntactic pivot (which is not the same as subject).

The hierarchy has held up remarkably well across forty-plus years of further typological work. It explains acquisition gradients (children acquire subject relatives before object relatives), processing complexity (object relatives are harder to parse than subject relatives in English, German, Mandarin), and cross-linguistic distribution. It is among typological linguistics' robust empirical generalisations.

Restrictive vs non-restrictive

English distinguishes two kinds of relative clause:

• Restrictive. Narrows the reference of the head noun. The students who failed retook the exam — only the failing students retook. No commas in writing; integrated intonation in speech.
• Non-restrictive (appositive). Adds information about an already-identified referent. The students, who failed, retook the exam — all the students failed and retook. Commas in writing; pause-bracketed in speech.

The grammatical reflex: non-restrictives require a wh-pronoun (which, who); that and bare relatives are restrictive only. *My brother, that lives in Berlin, called me is ungrammatical. The semantic difference is sharp — non-restrictive relatives are conventional implicatures (Potts 2005), contributing not-at-issue content that does not enter the at-issue assertion. Many languages (Mandarin, Japanese, Korean) do not grammaticalise the distinction; it surfaces only via context, intonation, or particles.

Variants and edge cases

• Free relatives. Headless relatives like what I bought = that which I bought. The head and the wh-pronoun fuse into a single element.
• Internally headed relatives. The head appears inside the relative clause — Quechua, Navajo, Lakhota. The head is interpreted twice: once inside the clause, once as the noun being modified.
• Correlatives. Hindi-Urdu, Sanskrit, Bambara, Hittite use a clause-initial demonstrative paired with a relative clause: jo larkii khaRii hai, vah merii bahin hai "REL girl standing is, that my sister is" = "the girl who is standing is my sister".
• Reduced relatives. English participial modifiers — the man eating the apple — derive historically from finite relative clauses (the man who is eating the apple). Their structure is debated.
• Stacked relatives. Multiple relative clauses on a single head: the book that I bought that you wanted to read. English allows recursive stacking; some languages cap it at one.

Common pitfalls

• Don't equate that-relatives with non-restrictives. The opposite is true: that-relatives are restrictive only; non-restrictives need a wh-pronoun.
• Don't confuse the relative pronoun with the interrogative pronoun. They share form but not function — relatives modify nouns, interrogatives form questions. Languages that distinguish them (Russian, Latin) use different morphology.
• Don't read the gap as covert in resumptive languages. Hebrew resumption is not a "filler-gap" with an invisible pronoun — the pronoun is overt and pronounceable.
• Don't assume Japanese relative clauses must be parsed top-down. Japanese is head-final; the head comes last. Successful processing relies on prediction and morphological cues, not waiting for the head.
• Don't confuse subject-gap with subject-of-relative. "The man who saw me" has a subject gap (relative clause subject is gapped). "The man I saw" has an object gap.
• Don't ignore the head-finality difference. English is head-initial within DPs (the relative follows the noun); Japanese is head-final (the relative precedes). The same Accessibility Hierarchy still applies — it is a hierarchy on positions, not orderings.