Carolyn Quam
My research asks how children learn to assign clearly discriminable phonetic variation to the appropriate levels of linguistic structure. Faced with highly variable acoustic input, how do young children converge on the correct interpretations of speech? In my dissertation work with Dr. Daniel Swingley, I focus on pitch as one particularly interesting dimension of variation. Pitch in English conveys meaning at many levels of structure; among other functions, it marks yes/no questions, helps cue lexical stress, and conveys pragmatic functions/emotional information. Unlike many languages, however, English does not use pitch to differentiate words. How do children figure out that English is not a tone language, while also discovering the relevant levels of pitch structure?
When do children learn that English does not have lexical tones?
Experiment 1: By 2.5 years, children know that English does not have lexical tones. We taught both adults and two-year-olds a novel word with a highly consistent, salient pitch contour (either a rise-fall or a low fall). The word was produced 22 times, first in a narration, then in ostensive labeling. In test, we changed either the pitch contour, the word’s vowel (from “deebo” to “dahbo”), or the initial consonant (to “teebo”). Adults looked less reliably at the target object when they heard “deebo” mispronounced as either “dahbo” (Quam & Swingley, in press, JML) or “teebo” (Quam & Swingley, 2007, BUCLD), but showed no sensitivity to the pitch change. Similarly, children responded to the change from “deebo” to “dahbo,” but not to the pitch change. (Download poster here: QuamSwingleyBUCLD07.pdf. Download manuscript here: Quam_Swingley_JML_InPress.pdf)
Experiment 2: Will younger children be more open-minded about treating pitch as lexical in English? We are using the Werker “Switch” procedure to investigate 18-month-olds’ interpretations of segmental versus pitch changes in a newly learned word. Children are habituated to two word-object pairings. Depending on the child, the words differ either in their vowel (“veedo” versus “vahdo”) or in their pitch contour (rise-fall versus low fall). We can then compare children’s looking time when they are presented with a word-object pair from habituation (“same” trials) versus a pair that violates the original pairings (“switch” trials). Preliminary evidence suggests that 18-month-olds are sensitive to both changes, suggesting that children constrain their hypotheses about lexical pitch between 18 and 30 months.
Can children exploit pitch regularity when it is relevant in English?
Experiment 3: Children do not exploit a pitch cue to the speaker’s emotions until age 4.5. After demonstrating that young children’s knowledge of English phonology enables them to ignore irrelevant pitch variation in word recognition, we next explore whether children can take advantage of pitch variation when it does convey relevant information. We use an interactive task to ask whether two- to five-year-old children can interpret a pitch cue to the speaker’s emotions. A Puppet, “Puppy,” expresses happiness or sadness using either body-language/facial cues or pitch. In the first version, children are asked to infer from Puppy’s emotions whether a particular toy is the one he was searching for. In the second version, children are simply asked to say whether Puppy is happy or sad (either by pointing to happy/sad faces or by saying “happy”/“sad”). In both cases, children do not succeed with the pitch cue until about age 4.5, though even two- and three-year-olds successfully exploit body-language/facial cues in the same task (Quam, Swingley, & Park, 2009). (Download poster here: SRCD09_Quam.pdf)
Experiment 4: When can children exploit a pitch cue to lexical stress? We test both adults’ and two- to five-year-olds’ ability to exploit the pitch cue to lexical stress in word recognition. Using “pitch resynthesis” in Praat, we manipulated tokens of “banana” and “bunny” to create correctly stressed (baNAna and BUnny) and misstressed (BAnana and buNNY) versions of each word that differed only in their pitch contour. We present participants with pictures of a bunny and a banana, and they hear either a word with correct stress (e.g., “baNAna”) or a misstressed word (e.g., “BAnana”). We can compare participants’ looking to the two pictures in response to the correct versus misstressed versions of the word, to see whether they look less at the target object when the pitch cue to stress is “mispronounced.” Preliminary evidence suggests that 4- and 5-year-olds are able to exploit the pitch cue to predict the word they are hearing, though their mispronunciation effect is smaller than that of adults. By contrast, 2- and 3-year-olds are not yet able to exploit the pitch cue in word recognition.
Does increased pitch variability improve children’s representations of newly learned words?
Experiment 5: Are children more sensitive to segmental mispronunciations of newly learned words if tokens of the word during teaching contain more acoustic variability? In a series of studies related to Experiment 1 above, we found that 24- and 30-month-olds did not respond to a change from “deebo” to “teebo.” In light of recent evidence that increased acoustic variability in training helps children identify the relevant variability during word recognition, we speculated that the consistency of the word’s pitch pattern during teaching may have hampered children’s detection of the consonant change. We are currently investigating whether children will detect the change from /d/ to /t/ if they are exposed to four different pitch contours during teaching.
How might the input convey to the learner that English does not contain lexical tones?
Finally, to complement my experimental work, I conducted an analysis of a corpus of recorded speech with Jiahong Yuan, a professor in Linguistics, and Daniel Swingley. We investigated how the pitch patterns of speech to young English learners might both convey the pitch categories of English (intonational categories like yes/no question vs. statement) and indicate that English is not a tone language. Using the Brent corpus from the CHILDES database (http://childes.psy.cmu.edu/), we found that words like "good" and "no" differ consistently from each other in their pitch characteristics across tokens. Despite this possibly misleading regularity, large within-word variability may convey to the child that English is not a tone language. I presented this work at CogSci 2008. (Download proceedings paper here: CogSci08quam_paper.pdf Download slides here: cogsci08quam_slides.pdf). We are currently developing a corpus of Mandarin speech to infants in order to compare the pitch characteristics of infant-directed Mandarin and English.
Current Research Projects
