Tracking Autocorrection to Explain its Sensitivity to AD
Grant
Overview
Affiliation
View All
Overview
description
Alzheimer’s disease (AD) is known to cause subtle changes in language production years before diagnosis, yet current behavioral tests take limited advantage of linguistic tools to diagnose AD at early stages. Most language tasks focus on single word production (e.g., picture naming), missing many critical aspects of language. Some tasks focus on connected speech (e.g., picture description), but they heavily rely on complex computational modeling methods, thus are hard to implement in clinical settings. Addressing these issues, autocorrection is a type of error produced in connected speech that is easy to elicit, easy to analyze, and is sensitive to AD biomarkers. Thus, the long-term goal of this proposal is to maximize this sensitivity, facilitating the development of the autocorrect task as a non-invasive, simple, and low-cost diagnostic tool for early detection of AD. Our proposed study will lay foundation to achieve this goal through investigating the underlying cognitive mechanisms that drive the sensitivity of autocorrection to AD. In the autocorrect task, participants read aloud short paragraphs in which some words are replaced by unexpected words that are similar in form (e.g., Think about they concept replaced the concept). Participants are told to read exactly what they see, but occasionally, they automatically correct the unexpected words and produce the expected words instead, i.e., they produce an autocorrect error (e.g., say the concept instead of they concept). Participants with AD or preclinical AD (i.e., those at risk for AD based on CSF biomarkers) produce more autocorrections than healthy controls, especially with function word targets with rich syntactic properties (e.g., the, and). This is probably because 1) AD decreases attention thus eliciting more misperception, 2) AD increases monitoring difficulty thus making it more difficult to detect planned errors, and/or 3) AD makes it more difficult to overcome competition from syntactically well-formed expected targets. We will combine behavioral and eye-tracking measures to test each account. In Aim 1, we will investigate the attention and monitoring accounts by manipulating the font color of autocorrect targets: they will be either in black or red font. We hypothesize that the red font will reduce sensitivity to AD, and skipping rate and regression rate in eye movements will reveal whether the effect is driven by facilitating attention or monitoring or both. In Aim 2, we will investigate the monitoring vs. syntactic constraints accounts. The autocorrect targets and their corresponding expected words will either match (e.g., much/more are both adverbs) or mis-match (e.g., then/that is an adverb/pronoun pair) in syntactic category. Mismatching pairs elicit greater syntactic anomaly than matching pairs, and thus should be harder to produce but be easier to monitor. This contrast, combining with regression rates in eye movements, will differentiate the syntactic constraints vs. monitoring account. The proposed project will be the first that combines the analysis of autocorrect errors with eye-tracking in preclinical AD, shedding light on how complex aspects of language production are affected by AD in its earliest stages in a simple way.
