Grosz et al. (2014)

Grosz et al. (2014) employed an experiment concerning the anaphoric properties of donkey sentences involving constraints on the antecedent. Three main constraints were investigated in the experiment – the overtness, the syntactic position and the uniqueness of an NP antecedent – to see what constrains the referentiality of the antecedent to the donkey

pronoun.

First of all, the overtness of the NP antecedent has been proposed to be crucial for a pronoun referring back to it, and it was the first constraint named the Overt NP Antecedent Constraint. Overtness is concerned with the noun phrase as an antecedent which could not be a part of a word, for example in the following examples extracted from Grosz et al. (2014:4), the antecedent fatherless in (15b) would be a non-overt antecedent while without a father in (15a) is an overt one.

(15) a. Every child who was without a father had lost him in the war.

b. Every child who was fatherless had lost him in the war.

Second, it was argued that the syntactic position of the antecedent would greatly influence the acceptability of referring a donkey pronoun back to its antecedent. This is named the Salient Position Condition, which states a predicate position is more salient than a modifier position. Below are examples taken from Grosz et al. (2014:4) with this Salient Position Condition, where in (16a) the predicate position is more salient than the modifier position in (16b).

(16) a. [Every child who was fatherless] had lost him in the war.

b. [Every fatherless child] had lost him in the war.

Furthermore, the third constraint was associated with the uniqueness of a singular donkey pronoun. It is greatly related to the world knowledge of the antecedent, for instance, while uniqueness is fulfilled in father since everyone can only have one biological father, it is

violated in friend since the presumption cannot be made regarding one would have only one friend. Through the experiment, Grosz et al. expected to see whether all three constraints would activate the referentiality of the antecedent to donkey pronouns, and whether the three interact with one another.

In addition to the above three main factors in the experiment, Grosz et al. discussed one more factor (i.e., the word type) – the N-less group and the N-owner group – since the Uniqueness Condition appears to apply as a different role in the two groups. The former group will not obtain the presupposition of the existence of the N while the latter will. For example, it is self-evident that horseless does not entail the subject possessing a horse; in contrast, horse-owner then certainly entails the subject having a horse.

Grosz et al. expected the results could support the theoretical analysis of donkey sentences; hence, the two competing approaches – the E-type pronoun approach and the dynamic semantic approach – were compared. The former approach discusses the existential reading of the sentence. On the other hand, the latter approach is concerned with how interpretations of sentences influence contexts; therefore, the process of interpretations is dynamic since new information in the discourse will keep updating old information⁷.

With regard to the item types of the antecedent, the approaches make different predictions on uniqueness. Concerning the N-less type, both the approaches predict that

7 One of the approaches to dynamic semantics is Discourse Representative Theory (DRT), which is what the unselective binding approach based on to be used to analyze Chinese conditionals. Under this approach, interpretations are discourse representatives that integrate together to offer complete information to discourse.

uniqueness plays a role in the interpretation of donkey pronouns. In terms of the N-owner type, the predictions of the two approaches diverge – the E-type approach still predicts the uniqueness of the antecedent, however, the dynamic approach does not.

A hundred and twenty-five adult subjects were recruited. An acceptability-rating task was conducted with a total of 102 sentences with 30 critical sentences and seventy-two fillers involved, and a plausibility norming task was employed consisting of thirty-nine pairs of test items. In the first task, the subjects were asked to choose the scale of five levels from

“extremely unnatural” to “extremely natural.” The four factors were taken into investigation – overtness, salience of syntactic position, uniqueness and word-types. Quantifiers in the experiment were every, no and many. In the norming task, it was to test the uniqueness of the antecedent.

The results showed four different effects. A prenominal-postnominal effect where the acceptability in the postnominal type was higher, indicated the salience of syntactic position did work. An effect of overtness showed the acceptability of the overt types greatly exceeded the other. An effect of word-type presented the N-owner type was much more acceptable than the N-less type, and an effect of uniqueness was obtained where the ratings in unique antecedents were higher than those in non-unique ones. Also, an interaction was obtained between uniqueness and the word-type where only the N-less type showed a uniqueness effect.

All in all, it was argued by Grosz et al. that the presences of the Overt NP Constraint and the Salient Position Condition were verified. Moreover, the interaction between uniqueness and word-type confirmed the dynamic approach, which predicted uniqueness would not operate in the N-owner type. This was the crucial evidence for Grosz et al. to propose that the dynamic approach to donkey sentences suited well.

The analysis and the design of Grosz et al.’s overall experiment were very novel and elegantly displayed, especially where they provided solid evidence to support the dynamic approach. Nonetheless, questions of subject recruitment and of the design in the two word-types arise. Since it was to see how the constraints interplay with referentiality, no contrasting group was compared but only one group recruited. As to the material design, it seems reasonable even without conducting an experiment to consider the N-less type has a uniqueness presupposition while the N-owner does not. Although the factor of the word-type seems appropriate, the contrast of the two designed types does not appear that valid. Hence, it might be more persuasive if the uniqueness effect could be found in all donkey sentence types instead of just the word-types that Grosz et al. designed. Also, the aim of the acceptability-rating task was designed to see the referentiality of the antecedent to the donkey pronoun; however, such a task may be challenging for the subjects to consider this issue.