Anaphora

A definite description (e.g., 'the car', 'the company') refers back to an entity previously introduced in the discourse, often via an indefinite noun phrase.

• Example: "I bought a car. The vehicle is red." — 'The vehicle' is coreferent with 'a car'.
• Bridging Inference: Sometimes the link is indirect. "I walked into the room. The ceiling was high." — 'The ceiling' is not directly coreferent but is inferentially associated with 'the room'.
• Lexical Variation: The anaphor often uses a synonym, hypernym, or descriptive phrase rather than repeating the exact antecedent.

A semantic relationship where a pronoun does not refer to a specific entity but acts as a logical variable bound by a quantifier.

• Example: "Every student thinks they will pass." — 'they' does not refer to a specific individual but co-varies with the set of students.
• Quantifier Scope: The interpretation depends on the scope of the quantifier ('every', 'each', 'no').
• Donkey Sentences: A complex case where an indefinite noun phrase acts like a universal quantifier: "If a farmer owns a donkey, he beats it."

A phenomenon in languages like Japanese, Spanish, and Italian where the anaphor is phonologically null—it is omitted entirely but semantically understood.

• Example (Spanish): "Juan llegó. ∅ Estaba cansado." (Juan arrived. [He] was tired.)
• Rich Agreement: Pro-drop is licensed by rich verbal inflection that marks person and number, allowing the subject to be recovered.
• Resolution Challenge: NLP systems must detect the empty syntactic position and resolve it to the correct antecedent, requiring syntactic parsing and discourse context.

An anaphoric reference that points not to a noun phrase but to an abstract entity, event, proposition, or fact described in a preceding clause or sentence.

• Example: "The team missed the deadline. This caused a major delay." — 'This' refers to the entire event of missing the deadline.
• Demonstrative Pronouns: Typically uses 'this', 'that', or 'it' to refer to propositions.
• Span Selection: Coreference systems must identify the correct textual span (often a clause or sentence) that represents the abstract referent, rather than a single noun phrase.

A scenario where a plural anaphor refers to multiple distinct entities introduced separately in the preceding discourse, requiring the merging of two or more antecedents.

• Example: "Alice met Bob at the conference. They decided to collaborate." — 'They' refers to the set {Alice, Bob}.
• Coordinate Resolution: The system must identify that the plural pronoun cannot be resolved to a single antecedent and must construct a combined referent.
• Agreement Features: The plural number feature on the anaphor is the primary signal that a split antecedent resolution is required.

The inverse of anaphora, where a pronoun or expression precedes the noun phrase it refers to. The reader must look forward to resolve the meaning.

• Example: "When he arrived home, John was exhausted."
• Technical Challenge: Requires models to maintain a look-ahead buffer and delay resolution, complicating left-to-right processing architectures.

The specific subtask of resolving pronouns (he, she, it, they) to their correct antecedent noun phrases.

• Key Features: Relies heavily on gender, number, and animacy agreement.
• Salience Models: Often use recency and grammatical role to rank candidate antecedents, as pronouns typically refer to the most prominent entity in the current discourse focus.

The complete ordered set of all mentions within a discourse that refer to a single entity.

• Structure: Links the first mention (the head) through all subsequent anaphoric references.
• Example: "Ada Lovelace wrote the first algorithm. She was a visionary. The Countess of Lovelace is celebrated today."
• Utility: Essential for entity-centric NLP tasks like knowledge base population and document summarization.

A phenomenon in pro-drop languages (e.g., Japanese, Spanish) where a syntactically omitted argument is semantically understood through coreference.

• Example (Spanish): "[Él] Comió la manzana. [Él] Estaba deliciosa." (He ate the apple. [It] was delicious.)
• Resolution: Requires the model to predict a null token and link it to an antecedent, demanding deep syntactic understanding beyond surface-form matching.

A non-identity relationship where a definite noun phrase refers to an entity inferentially linked to a previously introduced referent, rather than directly coreferring.

• Example: "We entered a restaurant. The waiter was rude."
• Mechanism: Relies on world knowledge and lexical semantics (part-whole, location-associate relations) rather than simple identity matching, making it a harder problem than standard anaphora.

A linguistic phenomenon where a demonstrative pronoun (this, that) refers to an abstract entity, event, or proposition described in a preceding clause or sentence.

• Example: "The team missed the deadline. This angered the client."
• Resolution Complexity: The antecedent is not a concrete noun phrase but a span of text or an event representation, requiring models to handle non-entity referents.