Recent & Upcoming Talks

In this talk, I present an overview of a lesser-known method, pupillometry, a natural, real-time measure of pupil dilation. Traditionally used to study cognitive response to memory, cognitive and attentional load, and emotional valence, pupillometry has also been used as a measure of effort expended in language comprehension. Increased pupil size has been observed for syntactically complex sentences (Engelhardt et al., 2010), violations of expected meter (Scheepers et al., 2013), and inadequate or misleading pitch accent (Zellin et al., 2011). I present a series of ongoing studies on classic attachment ambiguities in English, suggesting that the pupil responds to mismatches between prosodic and syntactic grouping. Preliminary analyses suggest that penalties for prosodic mismatch are also modulated by violation of additional structural or pragmatic expectations. I discuss implications for theories of prosody-syntax alignment in real-time sentence parsing, along with a comparison of the practical, methodological, and interpretive tradeoffs associated with this method.

A vast amount of literature suggests that the language processor generates expectations about upcoming material. Several studies have found evidence for a prediction error cost in cases where the comprehender encountered not the predicted word but a plausible unexpected continuation instead. This cost is argued to be a result of an inhibitory process that suppresses activation of the originally predicted word. Other studies, on the other hand, have found no such evidence for a prediction cost. In a probe recognition memory task, we find evidence for interference from an incorrectly predicted word, and in a self- paced reading study, we find evidence for facilitation when the originally predicted word is encountered later on in the sentence. Taken together, our results provide evidence against a strong version of the suppression account, in which all incorrectly predicted words are inhibited. Instead we argue in favor of a passive lingering activation account, in which activation for the disconfirmed prediction gradually decays over time.

In this paper we use pupillometry, a non-invasive, naturalistic method of measuring attention and cognitive load, to mea- sure the effect of stress clash (Chi.NESE ship) and its metrical repair (CHI.nese ship) during auditory sentence processing. We addressed two main research questions. The first question explores whether phonologically-disfavored metrical structure induce processing costs indexed by changes in pupil size.The second investigates whether the application of an optional pro- cess of stress retraction called the Rhythm Rule (Liberman & Prince, 1977) ameliorates or compounds any general penalty for stress clash. We find that unrepaired stress clash leads to greater pupil diameter relative to non-clashing sequences, indicating increased attention and cognitive load. We also find that repaired potentially-clashing sequences lead to a decrease in overall pupil diameter, indicating facilitation.

We introduce the LARC-ID, a pilot corpus of eye-movements obtained from subjects reading texts from a range of genres. Materials were presented in multiple paragraphs on the screen to more closely match naturalistic reading environments. Readers were encouraged to read for comprehension and enjoyment, engaging in various kinds of comprehension questions, including an open-ended reflection at the end of each text. Subjects also participated in a battery of individual difference measures, including those known to predict reading behavior in controlled experimental contexts, e.g., Rapid Automatized Naming (RAN; Denckla & Rudel, 1976), the author recognition (ART; Stanovich & West, 1989), and reading span (RSpan; Daneman & Carpenter, 1980) tasks. In addition to describing the central properties of the text and relationships between tasks in the battery, we present a sample analysis exploring how intrinsic lexical characteristics (length, frequency, and morphological complexity) interact with selected individual difference measures. The analysis provides the very first glimpse into what we hope will become a useful resource for reading researchers and educators. The entire corpus is freely available for unrestricted use.

Expressions such as ‘either’ can be used in a correlative structure (‘either…or’) or as a quantifier (‘either student’). Previous research found that sentence-initial correlative ‘either’ facilitated coordination processing by generating a prediction for clausal coordination. This prediction appeared even for quantificational uses. In the present study, a self-paced experiment found that correlative ‘both’ triggered a structural prediction sensitive to its syntactic position. In an eye-tracking experiment, however, there was no evidence of structural prediction when ‘both’ was unambiguously quantificational. We conclude that while correlatives generate predictions about upcoming structure, predictions across correlatives are not equally strong.

Cue-based retrieval models predict reduced retrieval accuracy when the retrieval cue overlaps with more than one item in memory (similarity-based interference, SBI). This study addresses the extent to which morphological gender and number interact with structural biases (Locality bias) during antecedent retrieval in Brazilian Portuguese sluiced sentences. The results of two self-paced reading studies indicate that both number and gender cues can generate SBI effects when the Locality bias is violated, corroborating previous findings (Harris, 2015, 2019). However, gender cues elicited earlier SBI effects than number cues did, suggesting that not all morphological cues have equal status in retrieval.

We report the results of a pupillometry experiment on the prosodic disambiguation of PP modifier sentences. We manipulated instrument plausibility (‘pen’ vs. ‘gun’) and prosodic boundary placement, so that a prosodic boundary biased towards a Modifier [A] or Instrument [B] construal of the prepositional phrase (‘The artist sketched [A] the man [B] with the pen/gun’). A growth curve analysis of change in pupil size suggests that listeners attend to prosodic boundary locations that disambiguate the structure, and that an online processing penalty, as indexed by increased pupil dilation, is generated when prosodic boundary information leads to an implausible construal.

Three visual-world eye-tracking experiments investigated how listeners responded to proper conventional (Exp.1) and unconventional (Exp.2-3) uses of contrastive accent. Exp.1 confirmed that conventional uses of contrastive accent lead anticipatory looks to a target object. Exp.2 showed no anticipation when contrastive accent was consistently misleading. When participants were told before the experiment that the speaker was not trustworthy in Exp.3, they learned to predict the upcoming referent with improper contrastive accent. The results suggest that listener’s beliefs about the speaker’s intentions modulates the degree of prosodic adaptation.

This study tested whether listeners adapt to speaker-specific prosody in anticipatory processing, and, if so, whether adaptation is modulated by belief about the speaker’s intention. In three visual-world eye-tracking experiments, we compared how listeners responded to proper uses of contrastive accent on an adjective-noun pair (e.g., First, find the red cat. Next, find the PURPLE_L+H* cat) in Experiment 1 to deviant or improper uses of contrastive accent (e.g., First, find the red cat. Next, find the PURPLE_L+H* pig) in Experiments 2 and 3. Experiment 1 confirmed that proper uses of contrastive accent lead anticipatory looks to the target object. Experiment 2 showed no anticipatory effects of contrastive accent when the adjective accent was deviant. Experiment 3, where participants were informed before the experiment that the speaker was not trustworthy, showed participants learned to anticipate the upcoming referent (pig) with improper contrastive accent over the course of the experiment.

Listeners must integrate multiple sources of information to construct an interpretation of a sentence. We concentrate here on the alignment of prosodic and syntactic grouping during online sentence comprehension. We present the results from a pupillometry study on the use of prosodic boundaries in resolving well-known attachment ambiguities. Using growth curve analyses to capture the non-linear dynamics of pupil dilation, we found increased pupil excursions for sentences that were disambiguated towards the dispreferred, non-local relation, especially when accompanied by supporting prosodic information. However, when prosodic and structural information did not align, pupillary response was muted, potentially indicating a failure to commit to a specific interpretation. More generally, the study shows how the currently under-utilized pupillometry method offers insights into spoken language comprehension.

Perspective shift allows speakers to utter content they may not fully endorse. We explore the interpretation of perspective shifting expressions with a study on Transparent Free Relatives (Allen poured [what (is called/he calls) a beergarita]) and orthographic Partial Quotation (Allen poured a ‘beergarita’) in a Speaker/Listener production-perception experiment. While PQ and TFRs with explicit subjects (what he calls) licensed shifted interpretations for the Speaker, only TFRs without explicit subjects licensed shifted interpretations for the Listener. We propose that PQ provides a less robust cue for perspective shift, even when the speaker attempts to prosodically mark it as non-speaker oriented.

We investigate how not-at-issue content is integrated into a sentence by comparing the online attachment of parenthetical relative clauses to that of better-studied restrictive relative clauses. In an eye-tracking experiment, we report an early penalty for high attachment in restrictive relative clauses (as predicted by processing principles such as late closure or recency) but not for parenthetical relative clauses. These findings suggest that the processor delays structural integration of not-at-issue material, even when attachment is disambiguated by verb agreement. A preliminary or underspecified parse may initially be computed for parenthetical clauses during comprehension.

Previous studies have observed a tendency to associate the remnant (e.g., who) of ambiguous sluicing ellipsis with the closest / most local correlate (someone) in the matrix clause, as in Somebody said Fred fired someone, but I don’t know who (Frazier & Clifton, 1998; Carlson et al., 2009; Harris, 2015). I present the results of two experiments investigating the interplay between locality and the discourse status of potential correlates. The studies exploit the discourse-linking property of which-phrases in ambiguous sluiced sentences, like A teacher scolded Max or Dotty, but I can’t remember which one, to explore whether the preference for more local correlates is modulated by the discourse status of the potential correlates. I propose a discourse economy constraint (Alternatives on Demand: Avoid positing new discourse alternatives without evidence), which interacts with structural constraints like locality. Evidence from several questionnaire studies, as well as two online self-paced reading studies, supports the predictions of a sentence processing model in which the discourse status of items in memory immediately impacts the retrieval of a correlate for the remnant of sluicing ellipsis.