Sensitivity to knowledge mismatch about identity in narrative

Sensitivity to knowledge mismatch about identity in narrative comprehension
Mindaugas Mozuraitis1,2, Craig G. Chambers1, Meredyth Daneman1
1
University of Toronto, 2Saarland University
Referential perspective-taking and “common ground” are not only essential for communication
in situated conversation but also for imagined worlds described in narrative discourse. Previous
work on narratives has shown comprehenders are able to make relatively subtle distinctions
between what they know vs. what various story characters are imputed to know based on
information provided in the narrative [1]. Furthermore, these distinctions are maintained across
longer narratives [2]. But are there limits to this ability in real-time processing? Work on realtime interactive conversation has shown that privileged knowledge about the identity of entities
is difficult for listeners to ignore compared to privileged knowledge about the presence of
entities in the current context [3]. Here we examine the ability to manage privileged knowledge
about identity vs. presence in narrative comprehension. This provides an important test of the
generality of earlier findings found in studies of conversation and also allows an evaluation of
whether listeners’ greater ability to suppress knowledge about object presence in interactive
conversation is related to the salience of the situational cues that often signal privileged
knowledge in these contexts, e.g., a physical occluder that hides an object from the view of the
conversational partner. In narrative, knowing and keeping track of whether the identity or
presence of an object is privileged or shared knowledge is based entirely on discourse memory.
We focus on comprehenders’ sensitivity to the fit between a linguistic description and a
referential entity in cases where this fit depends on knowledge attributed to a fictional “speaker”.
For example, a man who is a ranger could be referred to as “the ranger”, however, the felicity of
this description crucially depends on whether the fictional speaker knows the man’s occupation.
In Experiment 1, participants listened to 9 short recorded spoken critical narratives (randomly
intermixed with filler narratives), where the knowledge attributed to a protagonist was varied
relative to what was known to the listener-participant (see Example Narrative). There were three
experimental conditions: in control narratives, the protagonist was stated as being unaware of a
side character (a ranger). In referentially-transparent narratives, the protagonist was aware of
the side character as well as his/her occupation. In referentially-opaque narratives, the
protagonist was aware of the presence/existence of the side character but not his/her occupation.
Off-line comprehension questions given after each narrative confirmed that listeners could
correctly encode and remember the knowledge that should be attributed to the protagonist
regarding the presence or identity of the side character.
In Experiment 2, the same recordings were used in a look-and-listen paradigm that
allowed us to track listeners’ on-line referential hypotheses as target descriptions were heard in
Test Sentences. Critical displays (see Example Display) included a picture of a protagonist
(“Susan”), an unrelated object (coffee cup), a side character (ranger), and a target object
(raincoat). Critically, the target object’s name shared onset sounds with the name of the side
character’s occupation and was always mentioned in direct speech produced by the protagonist,
e.g., “Susan said to herself: I think the raincoat was a good idea given the weather”. As
predicted, the picture of the side character (ranger) attracted more eye fixations in the
referentially-transparent than in the control narratives (p<.05) as the target object name
(raincoat) unfolded in time. This confirmed that participants are sensitive to the protagonist’s
awareness of the presence/existence of the side character. Interestingly, in the referentiallyopaque narratives, the picture of the side character also strongly attracted early eye fixations, in
fact to the same extent as in the referentially-transparent narratives. This was the case even
though the protagonist could not plausibly be referring to the side character by occupation in the
referentially-opaque condition due to the protagonist’s ignorance of this information.
The results demonstrate that, consistent with recent findings from studies of conversation
[3], language users are quite ineffective at suppressing privileged knowledge about the identity of
entities during the early moments of processing of an unfolding noun—a result that contrasts
with the ability to overcome privileged knowledge about the contextual presence or existence of
entities. Importantly, this effect was found even though the way in which listeners became aware
of the shared vs. privileged status of knowledge was the same in both cases (i.e., information
from earlier narrative ), and despite the fact that off-line memory measures (Expt 1) showed that
comprehenders can reliably distinguish when identity information is privileged or shared.
Example Narrative
Introduction:
Control. Susan was enjoying a latte at her favourite café on a stormy day. Meanwhile, a ranger from
one of the national parks was helping track a bear that had escaped from the local zoo. The bear at that
moment was still eluding capture. However, Susan didn’t know anything about the bear or the ranger,
and was calmly sipping her coffee in the café.
Transparent. Susan was enjoying a latte at her favourite café on a stormy day. When she was there,
Susan would often see a tall man, who happened to be a ranger at one of the national parks. He was at
the café that afternoon as well. However, all Susan knew about him was that he was a ranger, and
nothing else beyond that, because they rarely talked.
Opaque. Susan was enjoying a latte at her favourite café on a stormy day. When she was there,
Susan would often see a tall man, who happened to be a ranger at one of the national parks. He was at
the café that afternoon as well. However, Susan didn’t know that he was a ranger, or anything else
about him, because they had never talked.
Test Sentence: As she finished the last sip of her latte, Susan said to herself “I think the raincoat I
brought today was a good choice given the weather”.
Conclusion: Then she ordered another latte and watched the storm outside turn the city streets into
small creeks.
Example Display
Side Character
Target Object
Unrelated Object
Protagonist
References
[1] Gerrig, R., Brennan, S., & Ohaeri, J. (2001). Jnl of Memory & Language, 44, 81-95.
[2] Lea, R., Mason, R., Albrecht, J., & Birch, S. (1998). Jnl of Memory & Language, 39, 70-84.
[3] Mozuraitis, M., Chambers, G. C., & Daneman, M., (2014). Talk given at the CUNY
Conference on Human Sentence Processing.