SRCD 2019: language input and prediction

Here are my slides for SRCD 2019.

A large body of research indicates that environmental factors like SES play a role in children’s language development (for review see Fernald & Weisleder, 2011). These findings reveal that children’s day-to-day language experiences vary tremendously, as do their developmental trajectories. Factors like SES, language input, and language processing abilities interact to shape children’s emerging linguistic abilities (e.g., Weisleder & Fernald, 2013).

At the same time, a related body of work has investigated whether a particular aspect of language processing – prediction – supports development, and a number of findings are in line with this view (e.g., Reuter et al., under revision). However, the origins of individual differences in prediction are unknown. Given prior findings linking language input and processing efficiency, it seems likely that language input also supports predictive language processing. Importantly, evaluating the link between input and prediction is essential for evaluating prediction-based theories of learning: These theories claim that accurate predictions are the result of accumulated language processing experiences. Children who receive more input have more processing experiences and therefore should be able to generate more accurate predictions during language processing (e.g., Dell & Chang, 2014).

We therefore hypothesized that home language input supports children’s emerging language processing abilities, including predictive language processing. Using a combination of eye-tracking tasks, LENA recordings, and vocabulary measures (PPVT), we find that:  High-SES toddlers receive marginally more language input from caregivers. High-input toddlers have larger vocabularies, and, importantly, high-input toddlers generate more robust predictions during language processing. These findings add to a body of literature linking SES-based disparities in input and processing (e.g., Weisleder & Fernald, 2013) and provide further support for prediction-based theories of language development (e.g., Dell & Chang, 2014).

We thank all participant families and we’re grateful to all Princeton Baby Lab research assistants for their help with data collection. We also thank Monica Ellwood-Lowe and Mahesh Srinivasan for organizing our SRCD 2019 symposium on SES-based disparities in language development.

SRCD 2019: prediction vs. repetition

Here is our poster for SRCD 2019.

Recent theories claim that prediction supports language development, and previous findings are in line with this view (Dell & Chang, 2014; Reuter et al., under revision). However, because behavioral measures of prediction are indirect, it’s unclear what representations are activated during processing, and how specific those representations are (Rabagliati et al., 2016). Is predicting a word like cookie via semantically related words like eat, yum, and mouth distinct from comprehending the word cookie itself? We used an eye-tracking paradigm to evaluate how infants processed different types of sentences (neutral, prediction, and repetition). Findings indicate that infants predict and comprehend words, replicating prior results (Reuter et al., under revision). Findings further suggest that infants can activate and pre-activate (predict) specific lexical representations during language processing: The behavioral dynamics of prediction are distinct from those of repeated comprehension. In sum, these findings further suggest that prediction supports both language processing and language development in infancy.

We thank all participant families. We also thank Chandra Greenberg and Claire Robertson for their assistance with stimuli and data collection, and Princeton Baby Lab research assistants for their help recruiting and scheduling participants.

SRCD 2019: fNIRS decoding

Here is our poster for SRCD 2019.

Verbal prediction is claimed to be an important mechanism supporting language development (Elman, 1990; Dell & Chang, 2014). In line with this view, prior findings indicate that infants can predict upcoming words during language processing (Reuter et al., under revision) and that prediction and vocabulary size are positively correlated (e.g., Mani & Huettig, 2012). However, previous research investigating verbal prediction has relied on behavioral methods and it’s unclear whether infants use top-down connections to pre-activate lexical representations. In two ongoing experiments, we extend prior fNIRS decoding methods (Emberson, Zinszer, et al., 2017) to decode infants’ lexical representations and to investigate whether infants are able to pre-activate (i.e., predict) lexical representations. These experiments make both methodological and theoretical contributions: First, it is possible to decode infants’ lexical representations using fNIRS decoding methods. Second, these methods provide a promising means of evaluating verbal prediction in infancy and determining to what extent infants pre-activate lexical representations. Future work will use fNIRS decoding to investigate verbal prediction among infants at-risk for language impairment in order to uncover the cognitive and neural origins of language delays.

We thank all participant families. We also thank Carolyn Mazzei and Jean Bellamy for their assistance with stimuli, and Princeton Baby Lab research assistants for their help recruiting and scheduling participants.

BUCLD 2018

Here are my slides for BUCLD 2018.

A number of recent theories propose that prediction facilitates efficient language processing. Supporting this idea are findings that listeners can use verb semantics and number markings to predict upcoming referents (Mani & Huettig, 2012; Lukyanenko & Fisher, 2016). However, precisely how prediction occurs during language processing remains uncertain. One prominent theory is prediction via simulation (Pickering & Garrod, 2013): Listeners use language production mechanisms to simulate the speaker’s upcoming production, which is contingent, at least in part, on perspective-taking and on well-developed language production mechanisms.

In the present study, we tested whether prediction occurs via simulation by evaluating whether listeners can use spatial deixis (this, that, these, and those) to predict the plurality and proximity of a speaker’s referent. In two eye-tracking tasks, English L1 adults, English L1 5-year-olds, and English L2 adults viewed scenes that included a speaker and four referents (experiment 1) or two referents (experiment 2). Participants listened to deictic sentences (e.g., Look at that wonderful cookie) and neutral sentences (e.g., Look at the wonderful cookie). Data collection for experiment 2 is ongoing, but preliminary findings suggest that only L1 adults are capable of prediction via simulation.

The present pattern of results suggests that prediction via simulation (Pickering & Garrod, 2013) supports processing for the mature, native speaker, but that extensive experience with cues in a language may be required before listeners can use this route for prediction. This three-group investigation goes beyond the empirical goal of assessing whether prediction occurs and evaluates how prediction occurs – a crucial goal for defining prediction’s role in language processing and learning.

We’re grateful to all participant families, to Claire Robertson for her assistance with stimuli, and to Mia Sullivan to her assistance with data collection and CHILDES coding.

she roars

This past week, I was honored to give a research talk at She Roars – a conference celebrating women at Princeton.

This conference was a unique opportunity to reflect on all the amazing women who have mentored me: Jenny Saffran, Maryellen MacDonald, Melanie Jones, Alexa Romberg, Jessica Willits, Jill Lany, Jess Hay, Jesse Snedeker, Manizeh Kahn, Melissa Kline, Lauren Emberson, Chris Potter, Elise Piazza… The list goes on and on! I’m so fortunate to be surrounded by all of these female scholars.

I left feeling a sense of camaraderie and inspiration, and can’t wait to return to Princeton for the next She Roars conference!