Replication Project: Preschoolers Flexibly Adapt to Linguistic Input in a Noisy Channel
Lydia Speyer
1 October 2017
1 Précis
Daniel Yurovsky’s, Sarah Case’s, and Michael C. Frank’s study on pre-schoolers’ adaptability to linguistic input in a noisy Channel was conducted in 2016 to find out whether preschoolers privilege either bottom-up cues from perceptual input or top-down expectations based on a speaker’s previous statement when presented with ambiguous sentences that could be interpreted in both ways or whether they flexibly adapt how much they rely on the different cues when the cues’ reliability changes. In their study, Yurovsky, Case and Frank conducted 3 separate experiments:
In the first experiment, 43 children aged 4 to 6 and 50 adults were asked to listen to a speaker who described one of two pictures. These picture pairs always included one semantically plausible and one semantically implausible picture. The spoken utterances were convolved with Brown noise of amplitude ~0.6 to increase the utterances’ ambiguity. Participants completed an exposure trial in which the referential expression accompanying the picture was unambiguous and a test trial in which the referential expression phonologically only differed in one sound and was therefore very ambiguous.
In the second experiment, Yurovsky, Case and Frank replicated the first experiment in a larger and developmentally broader sample. The participants were 140 children aged 3 to 5 who had to complete the same tasks as the children in experiment one.
The third experiment tested another prediction of noisy- channel processing: “As speech becomes noisier, and thus less reliable, children should rely more on their expectations” (Yurovsky, Case & Frank, 2016: 137). This experiment used again a similar design as the previous two experiments but it included two different versions of each acoustic recordings; one that was recorded in a sound proof room and one that was convolved with Brown noise of amplitude 0.7 which resulted in a signal-to-noise ratio of 19.5dB. The participants of experiment three were 114 children aged 4 to 5.
Overall, the results of the three experiments showed that, while children generally relied more on top-down expectations, they still, just like adults, flexibly adapted how much they relied on the different cues when the cues’ reliability changed.
2 Replication Project
2.1 Introduction
This study is a replication of Daniel Yurovsky’s, Sarah Case’s, and Michael C. Frank’s study on pre-schoolers’ adaptability to linguistic input in a noisy Channel. Conducted in 2016, Yurovsky, Case and Frank wanted to find out whether preschoolers privileged bottom-up cues from perceptual input or top-down expectations based on a speaker’s previous statement when presented with ambiguous sentences that could be interpreted in both ways. What they found was that children aged 4 to 5, just like adults, flexibly adapted how much they relied on the different cues when the cues’ reliability changed. In their study, Yurovsky, Case and Frank conducted 3 separate experiments; this replication project will only replicate experiment two.
2.2 Method
This section will give a detailed overview of the methods used in experiment two of Yurovsky’s, Case’s and Frank’s study. The replication project will follow these methods as closely as possible.
2.2.1 Participants
The participants of Yurovsky’s, Case’s and Frank’s noisy channel experiment were 146 children that were recruited from the floor of the Children’s Discovery Museum of San Jose, California. The children and their parents were approached by an experimenter who obtained informed consent before inviting them to participate in the study. 6 of the 146 children were later excluded as their parents indicated that the children were not exposed to English for more than 50% of the time. 48 of the children were 3 years old (24 boys, 24 girls), 49 children were 3 years old (19 boys, 30 girls) and 43 children were 5 years old (12 boys, 31 girls). While Yurovsky’s, Case’s and Frank’s participants were mostly exposed to American English (at least 50% of the time), this replication will be conducted in the United Kingdom; hence, this study will try to recruit native British children.
2.2.2 Stimuli, design, and procedure
The stimuli for the noisy channel experiment were several pairs of pictures that each were accompanied by a spoken utterance (speaker spoke American English). Each picture pair consisted of one semantically plausible picture and of one semantically implausible picture. The sentence accompanying the picture pairs described either the semantically plausible or the semantically implausible picture. All participants completed a block of eight exposure trials followed by a block of seven test trials. The order of trials was randomised across participants. In the exposure trial, the referential expressions for the two pictures were extremely distinct; hence the speaker’s reference was unambiguous. In the test trial, the referential expressions only differed phonetically in one sound, which made the accompanying utterances highly ambiguous. To further increase the ambiguity of the spoken utterances, the audio recordings were convolved with Brown noise of amplitude ~ 0.6 in the Audacity audio editor. The audio was transmitted through external computer speakers that were located approximately 2.5 feet away from the children. The pictures were created from freely available clipart.
Figure 1: Sample picture pair (exposure trial)
Figure 2: Sample picture pair (test trial)
Before starting the actual experiment, participants were introduced to the speaker - Katie. After seeing a picture of Katie, participants started with the experiment. In each trial, participants had to select the picture that they thought best matched the speaker’s description by touching one of the two pictures on an iPad. Participants were randomly divided into two groups: One group was exposed to a speaker that always selected the plausible picture in the exposure trials whereas the other group’s speaker always selected the implausible picture. In the test trials, the speaker referred to the implausible picture in both conditions.
This replication project will use the exact same pictures as the original experiment (freely available online: https://github.com/dyurovsky/noisy-kids/tree/master/stimuli/images) but will use the recordings of a British English speaker instead of an American English speaker as it is targeted towards British children and not towards American children. Other than that, the replication will follow the same methods as the original experiment.
2.3 Results
When analysing the data, Yurovsky, Case and Frank first fitted a separate mixed-effects logistic regression for each age group. They found that, in all age groups, children were more likely to choose the plausible referent when they were exposed to the speaker selecting the plausible referent in the exposure trials. To find out whether children’s behaviour changes with age, they then fitted a mixed-effects model to all the children’s data. The analysis showed that the older the children are, the more sensitive they react to the speaker’s choices in the exposure trial. Looking at the test trials, Yurovsky, Case and Frank found that while 4- and 5-year-olds used their previous experience with the speaker during the exposure trial to their advantage when interpreting ambiguous sentences, 3-year-olds did not; hence, they concluded that 3-year-olds do not yet use top-down expectations about a speaker’s intended meaning to process ambiguous sentences. To determine whether older children relied more on their expectations because they had built stronger expectations during exposure trial, Yurovsky, Case and Frank fitted an additional model that included the number of exposure trials on which individual children had selected the plausible picture. They found that older children did, in fact, form stronger expectations during exposure trials. Yurovsky, Case and Frank concluded that children aged 3 to 6 rely to a similar extent on expectations about a speaker’s intended meaning, but, across development, their ability to build these expectations gradually improves.
Figure 3: Results of test trial (experiment 2)
The replication project expects to find the same results as Yurovsky’s, Case’s and Frank’s study. This would provide further evidence that the noisy-channel principles are valid for both adults and children.
3 Bibliography
Yurovsky, D., Case, S., & Frank, M. (2016). Preschoolers Flexibly Adapt to Linguistic Input in a Noisy Channel. Psychological Science, 28(1), 132-140. http://dx.doi.org/10.1177/0956797616668557
Yurovsky, D. (2017). dyurovsky/noisy-kids. GitHub. Retrieved 2 October 2017, from https://github.com/dyurovsky/noisy-kids/tree/master/stimuli/images