Embedded Files
I'm interested in the social, interactional, and indexical functions of language, and in particular how this relates to the perception of speech. We believe that interactional and ethnographic and lab-based work are observing different facets of what is fundamentally the same thing--language use. I am interested in the ways that these facets fit together in an integrative view of perception and interaction.
My dissertation
My dissertation
Modeling the role of social information in speech perception
How do we model the relationship between “high level” social constructs and “low level” automatic processing of phonetic detail? Variation in pronunciation is socially informative, and listeners can draw on these social expectations when perceiving speech. This dissertation argues for a closer consideration of variation within sociophonetic exemplar modeling. I do this by reviewing the web of literature, simulating perception events in Python, and conducting an experiment. “Exemplar theory” is a class of models positing that past experiences interpreting stimuli are remembered as exemplars; new stimuli are categorized based on comparison to these stored memories. In particular, I focus on the Generalized Context Model (Nosofsky 1986; Johnson 1997), or GCM. The evidence that social categories, like other higher-order abstractions from stimuli, can play a role in categorization is well-established but loosely unified. Many adopt an episodic or exemplar-based framework in interpreting their results, but focus on the general patterns more than a specific model. I developed a Python library ExemPy which implements the GCM and provides routines for simulating common perception experiment tasks. I suggest applications for both enhancing empirical work and exploring theoretical space. I designed an experiment to explore a key difference among sociophonetic priming literature: whether social expectation is invoked as part of or outside of the phonetic stimulus. Taken together, this work advances an integrative, ecologically informed approach to exemplar-based sociophonetic research, drawing on multiple sources of evidence to contextualize our modeling.
You can read it here.
Implementing the generalized context model
Implementing the generalized context model
Exemplar modeling of spoken language perception can be leveraged in understanding behavioral data. The ExemPy library is a Python implementation of the Generalized Context Model (Nosofsky 1986; Johnson 1997, 2006) designed to make such modeling more accessible. ExemPy’s central use case is to categorize a set of stimuli based on a provided set of “exemplars.” The resulting dataframe resembles the aggregated data of a perception experiment.
Simulating perception gives us fine-tuned control over parameters and lets us visualize what perception would look like if different accounts are accurate. This can be compared to existing evidence or used to generate hypotheses for new work. There are also cases where a perception result would be useful, but isn’t possible to obtain. For example, when designing experiments, it can be helpful to know things like how similar two sounds are for listeners. If that research doesn’t yet exist, a simulation can provide some basis for the design.
The alpha version of the library is posted at https://github.com/emilyremirez/ExemPy.
Remirez_Johnson_LSA2021.mp4(Mis)Matching sociolinguistic cues: Evidence for interaction of phonetics & syntax in perception
(Mis)Matching sociolinguistic cues: Evidence for interaction of phonetics & syntax in perception
Remirez_LSA2020.pdfRemirez_ASA_F19.pdfThis project probes joint predictions of syntactic and phonetic exemplar theory. I'm also very interested in the relationship between high level social constructs and low level processing, and in the character of the unit of experience in these episodic models. Asking listeners if an enregistered construction sounds better in an accent enregistered to the same variety than in an unassociated accent will support our understanding of how much we can account for using episodic models of perception and what is stored with each episode. On average, listeners responded faster and rated sentences more favorably when the accent and syntactic structure of the stimulus 'matched,' or were more likely to have been experienced together previously. Unlike past research, this experiment found that for these listeners, it was African American English--not British English--that patterned with General American English. This could be due to confounds in the stimuli, the demands of the task, or language ideologies above the level of consciousness. Future work should examine on-line features more carefully.
Whispers formants & pitch: Evidence of abstraction
Whispers formants & pitch: Evidence of abstraction
This work was presented at the 2018 LSA meeting. Social information such as gender, ethnicity, or place of origin has been shown to affect speech perception when presented though visual stimuli or overt labeling. Because exemplar theory holds that traces prime other traces that contain similar information, this is typically interpreted in exemplar theoretic models as the socially-correlated stimulus increasing the activation level of congruous exemplars. However, in a speech perception model in which social information is stored alongside phonetic information, it is unclear whether this congruity happens purely at the level of basic phonetic information or at a more abstract level. This study explores that question using whispers.
Remirez_LSA.pdfParticipants were presented with a range of synthetic whispered words, based on natural recordings from a cisman and ciswoman, from an interpolated continuum of "male" to "female"sss formants. They were then tasked with selecting the 'speaker's' modally phonated voice from a synthesized continuum of male to female f0 (with the same formants). Results show an across-subjects correlation between low, narrow formants and low f0. Within-subjects, there is evidence of both a linear and non-linear relationship, the latter of which is consistent with abstraction and generalization based on social information. The task was completed both in the lab and on Amazon Mechanical Turk--you can play for yourself here!
map task, Ultrasound, & coronal deletion
map task, Ultrasound, & coronal deletion
MapTaskNoWPhon.pptxThis dataset, created in collaboration with graduate student Andrew Cheng, Professor Susan Lin, and undergraduate research assistants Sarah Chen and Mariya Rybak, was presented at Northwest Phonetics and Phonology (NoWPhon) 2017 and the Chicago Linguistics Society Workshop on Dynamical Systems. Familiar dyads complete a Map Task (Anderson et al. 1991)--a cooperative direction-giving game--while one participant is ultrasound imaged. Maps are designed to target deletion of coronal stops, while the nature of the task and relationship between participants encourage rapid, casual speech.
Pitch accent and information structure in south bolivian quechua
Pitch accent and information structure in south bolivian quechua
This project, born out of the 2016-2017 Field Methods class, taught by Professor Lev Michael, examined expression of focus and information structure categories through pitch accent in South Bolivian Quechua. The project involved elicitation, exercises from the Questionnaire on Information Structure (Skopeteas et al. 2006), and speech perception experiments. The methodology is discussed in a UC Berkeley Fieldwork Forum presentation. The experiments used stimuli from the speaker himself, some of which were resynthesized. This investigation concluded that while multiple types of focus are encoded through pitch accent, there is an asymmetry in production and perception.
"The DnD stuff"
"The DnD stuff"
Although I haven't worked in this area as a graduate student, many people are interested in my work on the fantasy role playing game Dungeons and Dragons. I video recorded a 1.5 hour game and selectively transcribed it using the Discourse Transcription conventions established by Du Bois et al. There were three sub-parts to analyzing this data set:
A Conversation Analysis style investigation of what I term 'reference alternation'--a player's ability to refer to themselves and others using their identity as a player, in "real life," or as their character within the keyed frame of the game. I find that ellipsis (the choice not to name a subject) is conditioned by the Dungeon Master being responsible for rolling the dice, rather than the player themself. Player identity is the default for these speakers, with character identity reference occurring in close proximity of each other. I analyze this as an instance of dialogic syntax (Du Bois 2014).
Analysis of gesture used to mutually co-construct the imagined landscape. I found that players do use similar gestures to organize space, but the topology is oriented around the ego, rather than in absolute terms. That is, each player would be consistent in placing a cave on their right, rather than consistent in placing it east.
A statistical analysis of coded qualitative data on reference alternation. Unfortunately, I can't locate the analysis, so the results are lost to time!
As some projects are on-going, I have not made all of my stimuli public. However, I will gladly share my stimuli and anonymized data privately by email request.
Engaging with research (or research pedagogy) from the other side of things!
Report abuse