Individual differences and contextual bias in pronoun resolution: Evidence from ERPs

Abstract

Although we usually have no trouble finding the right antecedent for a pronoun, the coreference relations between pronouns and antecedents in everyday language are often 'formally' ambiguous. But a pronoun is only really ambiguous if a reader or listener indeed perceives it to be ambiguous. Whether this is the case may depend on at least two factors: the language processing skills of an individual reader, and the contextual bias towards one particular referential interpretation. In the current study, we used event related brain potentials (ERPs) to explore how both these factors affect the resolution of referentially ambiguous pronouns. We compared ERPs elicited by formally ambiguous and non-ambiguous pronouns that were embedded in simple sentences (e.g., "Jennifer Lopez told Madonna that she had too much money."). Individual differences in language processing skills were assessed with the Reading Span task, while the contextual bias of each sentence (up to the critical pronoun) had been assessed in a referential cloze pretest. In line with earlier research, ambiguous pronouns elicited a sustained, frontal negative shift relative to non-ambiguous pronouns at the group-level.The size of this effect was correlated with Reading Span score, as well as with contextual bias. These results suggest that whether a reader perceives a formally ambiguous pronoun to be ambiguous is subtly co-determined by both individual language processing skills and contextual bias. IntroductionAlmost no conversation goes by without somebody using a pronoun to refer to some object, action or individual. Pronouns are our linguistic 'short-cuts' for maintaining reference to topics that are in the focus of our conversation. In fact, comprehending pronouns is such common practice, we usually feel as if we understand their antecedents immediately (e.g., Clark and Sengul, 1979). Interestingly, we generally do not even notice that the co-reference relations between pronouns and their antecedents in everyday language are often 'formally' ambiguous, that is, when the linguistic features of a pronoun (e.g., male/female, singular/ plural) by itself do not warrant the retrieval of a unique antecedent. For instance, given the sentence-fragment "Bruce Willis hated Al Pacino because he…", most people will automatically take 'Al Pacino' as the antecedent of 'he' (as the most likely continuation of this sentence will describe the characteristics of Al Pacino which he was hated for), even though 'he' might, in principle, also refer to 'Bruce Willis'. In

On sense and reference: Examining the functional neuroanatomy of referential processing

“…Because pronouns are among the most frequently used words in many languages, pronoun resolution is often regarded as the textbook example of referential processing. Consistent with our findings for nouns, referentially ambiguous pronouns (e.g., "Ronald told Frank that he…") also elicit a frontal negative shift compared to referentially coherent pronouns (Nieuwland and Van Berkum, 2006a;Van Berkum et al, 2004). Furthermore, referentially failing pronouns (when no suitable referent is directly available, e.g., "Rose told Emily that he…") elicit a P600 effect (Nieuwland and Van Berkum, 2006a;Van Berkum et al, 2004), suggesting that readers initially try to find an appropriate antecedent within the given set of discourse entities (e.g., Garnham, 2001) and therefore initially ascribe the referential failure to a problem with the syntactic gender of the pronoun (e.g., Osterhout and Mobley, 1995;see Van Berkum et al, 2007, for discussion).…”

“…1A and D). Because referentially ambiguous pronouns also elicit a sustained frontally-distributed negative shift in the ERP (e.g., Nieuwland and Van Berkum, 2006a), we speculate that at least some of the neural generators of this ERP effect are located in the reported dorsal and anterior/ ventral medial prefrontal regions. Although there is as yet no coherent account of the exact functional significance of these anterior prefrontal brain regions (BA10, e.g., Burgess et al, 2005;Ramnani and Owen, 2004), dorsal medial frontal regions have been implicated in making controlled, higher-order inferences to establish coherence during discourse comprehension (e.g., Ferstl and von Cramon, 2001), and anterior/ventral medial prefrontal regions have been related to making evaluative judgments (e.g., 5 Additional activations were located in the posterior cingulate gyrus (BA 31,24], voxel-level P = 0.065, voxel-level FDR corrected at P ≤ 0.05), middle temporal gyrus (BA 21,, voxel-level P = 0.085), and also right medial frontal gyrus (BA 8,[8,20,46], voxel-level P = 0.085).…”

“…To parallel natural reading times, all words (except from the word that directly preceded the critical pronouns onwards) were presented using a variable serial visual presentation procedure (see Nieuwland and Van Berkum, 2006a). Word duration was computed as ([number of letters * 30] + 190), with a maximum of 430 ms. From the word that preceded the critical pronoun onwards, every word was presented for a duration of 350 ms, with an exception for final words, which were presented 1000 ms. We adopted this presentation rate from an earlier ERP study (Nieuwland and Van Berkum, 2006a), where we used almost identical stimulus materials (as in that study, participants did not notice this alternation between variable and fixed word duration presentation within sentences).…”

“…Word duration was computed as ([number of letters * 30] + 190), with a maximum of 430 ms. From the word that preceded the critical pronoun onwards, every word was presented for a duration of 350 ms, with an exception for final words, which were presented 1000 ms. We adopted this presentation rate from an earlier ERP study (Nieuwland and Van Berkum, 2006a), where we used almost identical stimulus materials (as in that study, participants did not notice this alternation between variable and fixed word duration presentation within sentences). All inter-wordintervals were 150 ms. After every final word, a blank (bright) screen was presented for 2 s, which was subsequently followed by a fixation-mark.…”

“…180 sentences (a subset of 270 sentences used by Nieuwland and Van Berkum, 2006a) were created that each described the interaction between two characters. These two narrative characters were denoted by proper names (e.g., 'John/David/Mary/Lisa'), by definite noun phrases (e.g., 'the father/son/empress/queen') or they were celebrities (e.g., 'George Bush/Bill Clinton/Madonna/Jennifer Lopez').…”