21.4.2 Information Status
We noted above that the same referring expressions (such as many indefinite NPs) can be used to introduce new referents, while other expressions (such as many definite NPs, or pronouns) can be used to refer anaphorically to old referents. This idea of studying the way different referential forms are used to provide new or old information is called information status or information structure.
There are a variety of theories that express the relation between different types of referential form and the informativity or saliency of the referent in the discourse. For example, the givenness hierarchy (Gundel et al., 1993) is a scale representing six kinds of information status that different referring expression are used to signal:
The givenness hierarchy:
$$ \begin{array}{ll}in~focus>~activated>~familiar>~uniquely\quad&type\\\{it\}\quad&\left\{\begin{array}{l}that\\this\\ this~N\end{array}\right\}\quad\{\text{that}N\}\quad\{\text{the}N\}\quad\{\text{indef.~this}N\}\quad\{a N\}\end{array} $$
The related accessibility scale of Ariel (2001) is based on the idea that referents that are more salient will be easier for the hearer to call to mind, and hence can be referred to with less linguistic material. By contrast, less salient entities will need a longer and more explicit referring expression to help the hearer recover the referent. The following shows a sample scale going from low to high accessibility:
$$ \begin{aligned}&Full~name>long~definite~description>short~definite~description>last~name\\&>first~name>distal~demonstrative>proximate~demonstrative>NP>stressed\\&pronoun>unstressed~pronoun\\ \end{aligned} $$
Note that accessibility correlates with length, with less accessible NPs tending to be longer. Indeed, if we follow a coreference chain in a discourse, we will often find longer NPs (for example long definition descriptions with relative clauses) early in the discourse, and much shorter ones (for example pronouns) later in the discourse.
Another perspective, based on the work of (Prince, 1992), is to analyze information status in terms of two crosscutting dichotomies: hearer status and discourse status. The hearer status of a referent expresses whether it is previously known to the hearer, or whether it is new. The discourse status expresses whether the referent has been previously mentioned in the discourse.
The relationship between referring expression form and information status can be complicated; we summarize below three such complicating factors (the use of inferrables, generics, and non-referential forms):
Inferrables: In some cases, a referring expression does not refer to an entity that has been explicitly evoked in the text, but instead one that is inferentially related to an evoked entity. Such referents are called inferrables, bridging inferences, or mediated (Haviland and Clark, 1974; Prince, 1981; Nissim et al., 2004) Consider the expressions a door and the engine in sentence (21.46).
I almost bought a 1961 Ford Falcon today, but a door had a dent and the engine seemed noisy.
The indefinite noun phrase a door would normally introduce a new door into the discourse context, but in this case the hearer is to infer something more: that it is not just any door, but one of the doors of the Falcon. Similarly, the use of the definite noun phrase the engine normally presumes that an engine has been previously evoked or is otherwise uniquely identifiable. Here, no engine has been explicitly mentioned, but the hearer makes a bridging inference to infer that the referent is the engine of the previously mentioned Falcon.
Generics: Another kind of expression that does not refer back to an entity explicitly evoked in the text is generic reference. Consider example (21.47).
(21.47) I’m interested in buying a Mac laptop. They are very stylish.
Here, they refers, not to a particular laptop (or even a particular set of laptops), but instead to the class of Mac laptops in general. Similarly, the pronoun you can be used generically in the following example:
(21.48) In March in Boulder you have to wear a jacket.
Non-referential uses: Finally, some non-referential forms bear a confusing superficial resemblance to referring expressions. For example in addition to its referring usages, the word it can be used in pleonastic cases like it is raining, in idioms like hit it off, or in particular syntactic situations like clefts (21.49a) or extrapolation (21.49b):
(21.49) (a) It was Frodo who carried the ring.
(b) It was good that Frodo carried the ring.
21.5 FEATURES FOR PRONOMINAL ANAPHORA RESOLUTION
We now turn to the task of resolving pronominal reference. In general, this problem is formulated as follows. We are given a single pronoun (he, him, she, her, it, and sometimes they/them), together with the previous context. Our task is to find the antecedent of the pronoun in this context. We present three systems for this task; but first we summarize useful constraints on possible referents.
We begin with five relatively hard-and-fast morphosyntactic features that can be used to filter the set of possible referents: number, person, gender, and binding theory constraints.
Number Agreement: Referring expressions and their referents must agree in number; for English, this means distinguishing between singular and plural references. English she/her/he/him/his/it are singular, we/us/they/them are plural, and you is unspecified for number. Some illustrations of the constraints on number agreement:
John has a Ford Falcon. It is red. * John has a Ford Falcon. They are red.
John has three Ford Falcons. They are red. * John has three Ford Falcons. It is red.
We cannot always enforce a very strict grammatical notion of number agreement, since sometimes semantically plural entities can be referred to by either it or they:
(21.50) IBM announced a new machine translation product yesterday. They have been working on it for 20 years.
Person Agreement: English distinguishes between three forms of person: first, second, and third. The antecedent of a pronoun must agree with the pronoun in number. A first person pronoun (I, me, my) must have a first person antecedent (I, me, or my). A second person pronoun (you or your) must have a second person antecedent (you or your). A third person pronoun (he, she, they, him, her, them, his, her, their) must have a third person antecedent (one of the above or any other noun phrase).
Gender Agreement: References also must agree with the gender specified by the referring expression. English third person pronouns distinguish between male, (he, him, his), female, (she, her) and nonpersonal (it) genders. Unlike in some languages, English male and female pronoun genders only apply to animate entities; inanimate entities are always nonpersonal/neuter. Some examples:
(21.51) John has a Ford. He is attractive. (he=John, not the Ford)
(21.52) John has a Ford. It is attractive. (it=the Ford, not John)
(21.53) John bought himself a new Ford. [himself=John]
Binding Theory Constraints: Reference relations may also be constrained by the syntactic relationships between a referential expression and a possible antecedent noun phrase when both occur in the same sentence. For instance, the pronouns in all of the following sentences are subject to the constraints indicated in brackets.
(21.54) John bought him a new Ford. [him\neq John]
(21.55) John said that Bill bought him a new Ford. [him≠Bill]
(21.56) John said that Bill bought himself a new Ford. [himself=Bill]
(21.57) He said that he bought John a new Ford. [He\neq John; he\neq John]
English pronouns such as himself, herself, and themselves are called reflexives. Oversimplifying the situation, a reflexive corefers with the subject of the most immediate clause that contains it (ex. 21.53), whereas a nonreflexive cannot corefer with this subject (ex. 21.54). That this rule applies only for the subject of the most immediate clause is shown by examples (21.55) and (21.56), in which the opposite reference pattern is manifest between the pronoun and the subject of the higher sentence. On the other hand, a full noun phrase like John cannot corefer with the subject of the most immediate clause nor with a higher-level subject (ex. 21.57).
These constraints are often called the binding theory (Chomsky, 1981), and quite complicated versions of these constraints have been proposed. A complete statement of the constraints requires reference to semantic and other factors, and cannot be stated purely in terms of syntactic configuration. Nonetheless, for the algorithms discussed later in this chapter we will assume a simple syntactic account of restrictions on intrasentential coreference.
Selectional Restrictions: The selectional restrictions that a verb places on its arguments (see Ch. 19) may be responsible for eliminating referents, as in example (21.58).
John parked his car in the garage after driving it around for hours.
There are two possible referents for it, the car and the garage. The verb drive, however, requires that its direct object denote something that can be driven, such as a car, truck, or bus, but not a garage. Thus, the fact that the pronoun appears as the object of drive restricts the set of possible referents to the car. Selectional restrictions can be implemented by storing a dictionary of probabilistic dependencies between the verb associated with the pronoun and the potential referent.
Recency: We next turn to features for predicting the referent of a pronoun that are less hard-and-fast. Entities introduced in recent utterances tend to be more salient than those introduced from utterances further back. Thus, in example (21.59), the pronoun it is more likely to refer to Jim's map than the doctor's map.
(21.59) The doctor found an old map in the captain's chest. Jim found an even older map hidden on the shelf. It described an island.
Grammatical Role: Many theories specify a salience hierarchy of entities that is ordered by the grammatical position of the expressions which denote them. These typically treat entities mentioned in subject position as more salient than those in object position, which are in turn more salient than those mentioned in subsequent positions.
Passages such as (21.60) and (21.61) lend support for such a hierarchy. Although the first sentence in each case expresses roughly the same propositional content, the preferred referent for the pronoun he varies with the subject in each case – John in (21.60) and Bill in (21.61).
(21.60) Billy Bones went to the bar with Jim Hawkins. He called for a glass of rum. [he = Billy]
(21.61) Jim Hawkins went to the bar with Billy Bones. He called for a glass of rum.
[he = Jim]
Repeated Mention: Some theories incorporate the idea that entities that have been focused on in the prior discourse are more likely to continue to be focused on in subsequent discourse, and hence references to them are more likely to be pronominalized. For instance, whereas the pronoun in example (21.61) has Jim as its preferred interpretation, the pronoun in the final sentence of example (21.62) may be more likely to refer to Billy Bones.
(21.62) Billy Bones had been thinking about a glass of rum ever since the pirate ship docked. He hobbled over to the Old Parrot bar. Jim Hawkins went with him. He called for a glass of rum. [ he = Billy ]
Parallelism: There are also strong preferences that appear to be induced by parallelism effects, as in example (21.63).
(21.63) Long John Silver went with Jim to the Old Parrot. Billy Bones went with him to the Old Anchor Inn. [ him = Jim ]
The grammatical role hierarchy described above ranks Long John Silver as more salient than Jim, and thus should be the preferred referent of him. Furthermore, there is no semantic reason that Long John Silver cannot be the referent. Nonetheless, him is instead understood to refer to Jim.
Verb Semantics Certain verbs appear to place a semantically-oriented emphasis on one of their argument positions, which can have the effect of biasing the manner in which subsequent pronouns are interpreted. Compare sentences (21.64) and (21.65).
(21.64) John telephoned Bill. He lost the laptop.
(21.65) John criticized Bill. He lost the laptop.
These examples differ only in the verb used in the first sentence, yet the subject pronoun in passage (21.64) is typically resolved to John, whereas the pronoun in passage (21.65) is resolved to Bill. It has been argued that this effect results from what the “implicit causality” of a verb: the implicit cause of a “criticizing” event is considered to be its object, whereas the implicit cause of a “telephoning” event is considered to be its subject. This emphasis results in a higher degree of salience for the entity in this argument position.