Binding Theory

Three nominal classes, three principles

Why does John_i saw himself_i sound fine while *John_i saw him_i (with him referring to John) sound contradictory? And why does John_i believes Mary saw him_i reverse the pattern, with the pronoun coreferring with John but the reflexive failing? Binding theory's answer is that the grammar partitions nominals into three classes, each subject to a different structural rule.

Class	Examples	Principle	Statement
Anaphor	himself, herself, itself, themselves, each other, oneself	A	Must be bound in its governing category
Pronoun	he, she, it, they, him, her, them, his, her	B	Must be free in its governing category
R-expression	John, Mary, the king of France, the woman next door	C	Must be free everywhere

The terms bound and free are technical: a node is bound by an antecedent that c-commands it and is co-indexed; otherwise it is free. The governing category is roughly the smallest clause containing the nominal, its governor (the verb or preposition that case-marks it), and an accessible subject. The principles produce a complementary distribution: where Principle A demands a local binder for an anaphor, Principle B forbids one for a pronoun.

C-command and the governing category

C-command, introduced by Tanya Reinhart's 1976 MIT dissertation, is the structural relation that makes binding theory work. A node A c-commands a node B if the first branching node dominating A also dominates B, and A does not dominate B. Intuitively, A c-commands its sister and everything inside her, but not her ancestors.

The relation is purely structural — linear order is irrelevant. John's mother saw himself is ungrammatical not because John is too far from himself linearly but because John sits inside the subject DP and fails to c-command the reflexive. The c-commanding nominal is the entire DP John's mother, which is feminine and gender-clashes.

The governing category is the binding-theoretic locality domain. For most cases it is the smallest clause containing the nominal, its governor, and an accessible subject. Three sample governing categories:

• John saw [himself in the mirror]. — Governing category for himself: the matrix clause. Subject John binds the anaphor. Principle A satisfied.
• John thinks [Mary saw himself]. — Governing category for himself: the embedded clause. The only c-commanding subject inside is Mary, gender-clash, no licit binder. Principle A violated.
• John believes [himself to be smart]. — Governing category extends to the matrix clause because the embedded infinitival has no accessible subject (ECM construction). John binds himself. Principle A satisfied.

Worked examples

Binding theory's predictions become vivid when we walk through a small inventory of test sentences. Indices show intended coreference; * marks ungrammaticality on the indicated reading.

Example	Principle in play	Verdict
Johni saw himselfi.	A: anaphor bound locally	Grammatical
*Johni saw himi.	B: pronoun bound locally	Ungrammatical
Johni said that Mary saw himi.	B: pronoun free in its clause	Grammatical
*Johni said that Mary saw himselfi.	A: anaphor not bound locally	Ungrammatical
*Hei thinks that Johni is smart.	C: R-expression c-commanded	Ungrammatical
Johni's motherj saw himi.	B: John doesn't c-command him	Grammatical
Johni believes himselfi to be smart.	A: ECM extends governing category	Grammatical
*Johni believes himselfi is smart.	A: finite clause blocks ECM	Ungrammatical

The fifth row is Principle C in action. The pronoun He c-commands the R-expression John (it's the matrix subject), so co-indexing with John would mean John is bound — but R-expressions must be free everywhere. The same configuration with John in subject position and He further down is fine — Principle C sees only c-command, not linear order.

ECM and the believe-want-expect class

The contrast between John believes himself to be smart and *John believes himself is smart is one of binding theory's cleanest empirical wins. The verbs believe, expect, want are exceptional case markers (ECM): they take an infinitival complement whose subject is case-licensed by the matrix verb across the clause boundary. Because the embedded clause is infinitival and its subject is case-licensed externally, it lacks the accessible subject that would make it a governing category. The governing category therefore extends to the matrix clause, where John can bind himself.

If we make the embedded clause finite — believes himself is smart — the embedded clause regains an accessible subject (himself itself, in subject position) and becomes its own governing category. Now himself needs a binder inside that clause, but its only c-commanding nominal is itself. Principle A is violated; the sentence is ungrammatical.

The same logic predicts the contrast for pronouns: ?John believes him to be smart (with him = John) is bad because the matrix-extended governing category now contains a c-commanding John, violating Principle B. John believes he is smart (with he = John) is fine because the embedded finite clause is the governing category and the pronoun is free in it.

Cross-linguistic variation

The inventory of anaphors and pronouns varies sharply across languages, even though the principles are claimed to be universal.

Language	Anaphor inventory	Locality of anaphor
English	himself, each other	Strictly local (clause-bound)
Norwegian	seg-selv (local), seg (long-distance)	Two-way split
Icelandic	sjálfan sig (local), sig (long-distance, logophoric)	Two-way split + logophoric uses
Mandarin	ziji	Long-distance, even across multiple clauses
Latin	se	Long-distance into infinitival/subjunctive complements
Japanese	zibun	Long-distance, often discourse-bound

Mandarin ziji is the classic challenge. Zhangsan rending Lisi piping-le ziji ("Zhangsan thinks Lisi criticised ziji") admits both a local reading (ziji = Lisi) and a long-distance reading (ziji = Zhangsan). Cole, Hermon, & Sung (1990) argued that long-distance ziji moves covertly at LF, picking up successive subjects until it lands at its antecedent. Reinhart & Reuland's 1993 reformulation distinguishes SE-anaphors (simple, like seg, sig, ziji) from SELF-anaphors (compound, like himself, seg-selv): only SELF-anaphors must be locally bound; SE-anaphors permit long-distance binding through logophoricity.

Puzzles and revisions

Classical binding theory has known holes. Picture-NP reflexives appear without an obvious local binder — The portrait of himself in the morning paper made John uncomfortable. Logophoric reflexives in Icelandic and African languages (e.g. Ewe, Yoruba) take antecedents that are perspective-holders, not c-commanding subjects. Reciprocals (each other) require plural antecedents, complicating the simple binding statement.

Reinhart & Reuland (1993) "Reflexivity" reformulated binding by separating reflexivisation (a property of predicates: a predicate is reflexive if two of its arguments are coindexed) from chain formation (the structural condition). Their Condition A says a reflexive-marked predicate is reflexive; Condition B says a non-reflexive predicate is non-reflexive. The reformulation handles long-distance anaphora and picture-NP reflexives without stipulating a special governing category, but at the cost of more abstract machinery.

Other revisions: Pollard & Sag (1992, 1994) within HPSG; Hellan (1988) for Norwegian; Heim (1998) on the semantics of binding; Heim & Kratzer (1998) reformulating binding within compositional semantics. The empirical generalisations are robust; the precise statement of the locality domain is contested.

Variants and adjacent constructions

• Reciprocals. Each other, one another behave like anaphors but require a plural antecedent and obey reciprocal interpretation rules (Heim, Lasnik, & May 1991).
• Logophors. Reflexives that pick up a perspective-holder antecedent rather than a c-commanding one. Common in Icelandic, Ewe, Japanese.
• Picture-NP reflexives. That picture of himself surprised John. The reflexive is inside a DP whose head doesn't take a subject; binding theory must be relaxed.
• Inalienable possession. John raised his hand with possessive his = John is bound only loosely; some languages use a reflexive (Slavic).
• Control PRO. The silent subject of an infinitive — John tried [PRO to leave] — is governed by Control Theory, sister to binding theory.

Common pitfalls

• Don't confuse c-command with linear precedence. John's mother saw himself fails because John doesn't c-command, even though it precedes the reflexive.
• Don't read Principle C as forbidding all repeated names. John thinks John is smart is fine if the two Johns are different people; Principle C only forbids c-commanded coreference.
• Don't apply English Principle A to Mandarin ziji. The locality domain varies cross-linguistically; long-distance binding is widespread.
• Don't ignore the difference between binding and coreference. Two NPs can be coreferential without one binding the other (no c-command). Heim's "Coreference Rule" handles the residue.
• Don't take "free in governing category" to mean "no antecedent at all". The pronoun can have an antecedent — just not a c-commanding one inside the governing category.