LANGUAGE PRODUCTION UNDER UNCERTAINTY
One line of my work focuses on language production under uncertainty: asking if, how, and when speakers begin planning their utterances when they are still uncertain about the message they would like to convey.
To start answering these questions, I developed a novel picture-description paradigm that manipulates how much information participants have about their message, and consequently, how much of their utterance they can plan. In a preliminary study, we found that uncertainty affects speakers’ word order, as they prioritize message components that are certain to be useful and place them first in the utterance (Gussow & MacDonald, 2023). Moreover, this strategy allows speakers to begin production sooner, and they can continue planning the rest of the utterance incrementally, as more information becomes available.
We next examined production strategies in describing ongoing events (Gussow & MacDonald, 2026), inspired by sportscasters who narrate rapid, unpredictable events in real-time. Participants described videos of motion events (e.g., a butterfly flying towards a slide), and we manipulated whether the goal (e.g., the slide) was known from the start of the trial, or ambiguous until the event concluded. We found that participants began speaking with limited message information – while the event was ongoing and its outcome still uncertain. Although this strategy allowed them to begin speaking sooner, it resulted in longer word durations and more silent pauses, as they planned upcoming material while speaking. Moreover, in a cross-linguistic comparison between Spanish and English, we found that speakers adapt their strategies to the word order of the language: in English, the uncertain part of the message had to be produced early in the utterance, causing later response times and more disfluencies; while in Spanish it could be postponed – allowing speakers to begin speaking early and fluently even when uncertainty remains.
My most recent project investigates how speakers’ planning strategies change across the adult lifespan. Prior research on production in older adults generally shows difficulties with word retrieval and articulation, resulting in higher rates of disfluency and slower speech. But what happens when the message is uncertain? Difficulty at the message stage might simply add to these age-related effects, or it could disproportionately affect older adults, changing how they plan under uncertainty. Working with undergraduate summer fellows at the University of Richmond, we tested this by creating images of scenes that were either concrete – with easily identifiable messages; or abstract – where messages were difficult to identify. We recorded younger (20-35) and older adults (60-75) describing these simple scenes (e.g., a woman is reading). Results show slower speech rates and later onset times among older adults and in the abstract message condition, but no interaction emerged between age and condition. This suggests that despite general age-related slowing, older adults remain similarly sensitive to message uncertainty as younger adults. As a follow-up, we plan to introduce greater time pressure, forcing message processing to overlap with word retrieval and articulation. Age-related differences in planning strategies may emerge under these more demanding conditions, which also better reflect fast-paced everyday conversation. By distinguishing general cognitive slowing from changes in language planning itself, this work can help clarify which aspects of language production remain stable throughout adulthood and how the language system adapts to age-related changes in broader cognition.
PLANNING PARALLELS IN LANGUAGE AND MOTOR ACTION
Language and action are typically studied separately, but there are several parallels between the two domains, particularly because both require a hierarchical plan and sequencing of sub-goals to reach an overarching goal. More generally, I view both language and action as continuous decision-making processes, as producers choose between multiple behavior options – whether sentences and words or trajectories and grasps – to achieve their goal.
This project started as a cross-university collaboration with interdisciplinary researchers from psycholinguistics, comparative psychology, and kinesiology. Together, we developed parallel tasks in language and motor domains, and compared participants’ performance to uncover similarities in cognitive plans. For example, we showed how producers in both domains reuse components of prior plans when planning new ones (Lebkuecher, Schwob, Kabasa, Gussow, MacDonald, & Weiss, 2022), and that structurally analogous sequences designed to be difficult (“Tongue Twisters” in language and “Finger Fumblers” in action) yield similar error patterns across domains (Gussow, Weiss, & MacDonald, 2023). Findings from these comparisons not only benefit each domain individually, but also contribute to our understanding of domain-general cognitive planning principles.
PREDICTIVE
COMPREHENSION
In prior work during my Master’s degree, I used eye-tracking in the visual world paradigm to investigate predictive processing during Spanish sentence comprehension. We found that although comprehenders initially use the semantic context of a sentence to predict upcoming words, they can rapidly adjust their predictions based on grammatical gender cues. However, the sentence context still affects integration of the target word, suggesting a lingering effect of the initial prediction (Gussow et al., 2019). I later continued this line of work with Yuzhe Gu, my undergraduate mentee at UW-Madison, whose senior thesis used a mouse-tracking paradigm to show that native English speakers use phonological cues (a versus an) to predict upcoming words, but nonnative speakers (L1 Mandarin–L2 English) are less sensitive to these cues. Predictive processing is now central to language comprehension models, making it important to understand how it varies with language experience. I also view predictive comprehension as the parallel of production under uncertainty, tying these lines of my work together: in both cases, we use partial information to get a head start on processing, making it more efficient yet error-prone.