December 07, 2006

Inherent ambiguity as an adaptive advantage

I tend to be suspicious of new theories advanced by people who want to sell me things. There's a good reason why "federal law ... states that intramural scientists conducting research with human subjects—for example, investigators and research team members involved in patient selection, the informed consent process, and clinical management of a trial—should not be allowed to have any financial interest in or relationship with any company whose interests could be affected by their research or clinical trial..."

I'm especially suspicious when the new theories are described only in vague terms, and no evidence is offered other than even vaguer mentions of large numbers of satisfied customers, and references to authoritative places where unspecified research might have been done.

Priscilla Dunstan's theory about the secret language of babies, documented on Oprah ("Amazing Medical Breakthroughs: The Secret Language of Babies Video") and available to parents and other interested parties in a 2-DVD set for $59.95, starts out with all of these handicaps. That doesn't mean that it's wrong. Time will tell. But as I'll explain below, I think that even if it's nonsense, it's probably beneficent nonsense. Unless you think that using false or exaggerated claims to sell things is intrinsically wrong -- and in that case, where would civilization be?

The first new aspect of Dunstan's idea, as far as I can tell, is that it's important to distinguish between a "pre-cry" -- which seems to be a sort of crying, but one occurring early in the time course of a crying bout -- and the "hysterical cry", which is said to follow if the "pre-cry" is not attended to. In addition, there's a specific lexicon of five "words" that are claimed to be used in different types of pre-cry, which Ms. Dunstan describes as follows (this version of the list is taken from Oprah's web site):

  • Neh="I'm hungry"
  • Owh="I'm sleepy"
  • Heh="I'm experiencing discomfort"
  • Eair="I have lower gas"
  • Eh="I need to burp"

Following yesterday's appeal, a reader is lending me a copy of the Dunstan DVD set, and I hope to learn more from it exactly what Ms. Dunstan thinks these five pre-cry "words" are like, in acoustic terms. Meanwhile, I'll explain a bit about past research on infant crying.

There's a long history of acoustic studies of infant cries, going back at least to the 1960s -- I recall learning about this from Eric Lenneberg in a course I took as an undergraduate, about 40 years ago. A sample reference would be O. Wasz-Höckert, J. Lind, V. Vuorenkoski, T. Partanen and E. Valanne, "The infant cry: A spectrographic and auditory analysis", Vol. 29 of Clinics in Developmental Medicine (Spastics International Medical), 1968.

A summary of four decades of back-and-forth on this topic can be found in Joseph Soltis, "The signal functions of early infant crying", Behavioral and Brain Sciences 27, 443-490, 2004:

Infant crying is clearly a means by which infants can communicate needs (e.g., hunger, pain, or discomfort) to caregivers, who may be alerted to appropriately satisfy those needs (e.g., by feeding, protecting, or soothing). It is a matter of some controversy, however, as to whether there are acoustically distinct cry types (e.g., hunger cries or pain cries) to which caregivers can respond specifically without additional contextual cues (reviewed by Gustafson et al. 2000).

Work in the 1960s by the so-called Scandinavian cry group is often cited in support of the cry types hypothesis (see Gustafson et al. 2000; Wasz-Hockert et al. 1985). Researchers recorded birth cries, pain cries (during vaccination), hunger cries (4 hours after feeding), and pleasure cries (after feeding). Listener subjects identified the four cry types better than chance (55% correct versus 25% expected). Gustafson et al. (2000) criticized this work, however, because the best exemplars of each cry type were preselected by researchers, and listeners were given the four a priori categories in advance, both of which conditions increased the likelihood of accurate assignment. Additionally, the results were collapsed across all four cry types, so that the positive effect could have been due to only the most easily distinguishable cries, such as the contented coos and babbles that constitute the “pleasure cry.” In a replication of earlier work, however, exemplars for each cry type were chosen at random, and results for the four cry types were presented separately (Wasz-Hockert et al. 1968). Additionally, the replication showed that the four cry types differed statistically along several acoustic dimensions, such as fundamental frequency and melody, although the differences were quantitative rather than qualitative. The accuracies in identifying birth, pain, hunger, and pleasure cries were 48%, 63%, 68%, and 85%, respectively (grand mean = 66%).

Other studies also show that subjects can distinguish between cry types. Wiesenfeld et al. (1981) showed that mothers could identify pain cries (rubber band snap) versus anger cries (taking away pacifier or physical restraint) of their own infants better than chance when given three categories from which to choose (pain, anger, or other; 66% correct versus 33% expected). Gustafson et al. (2000) also showed that mothers could identify pain versus hunger cries better than chance when given six cry categories from which to choose (44% correct vs. 17% expected).

When subjects are given open-choice tests, however, the evidence for cry types is not as strong. In an early study by Sherman (1927), cries were elicited from babies by late feeding (hunger cry) and by dropping, restraining, or pricking with a pin (pain cries). Non-mother subjects behind a screen were asked to judge the “emotional characteristics” of the cries, but there was no agreement among listeners, leading to the conclusion that infant cries were merely “undifferentiated noise” (Gustafson et al. 2000). More recently, Muller et al. (1974) played cries elicited by hunger (pulling the nipple away), pain (rubber band snap), and startling (clap of wooden blocks close to the ears). Again, without a priori categories, subjects could not differentiate the cry types, even of their own children. Participants tended to attribute all cries to hunger.

An alternative to the cry types model is the view of the infant cry as a graded signal (Gustafson et al. 2000; Murray 1979). According to this view, vocalizations vary quantitatively on some acoustic dimension, such as duration or frequency, and that graded change along the dimension reflects motivational or emotional state.

(Gustafson 2000, often cited in this passage, is G.E. Gustafson, R.M. Wood and J.A. Green, "Can we hear the causes of infants' crying? In: R.G. Barr et al., Eds., Cry as a sign, a symptom and a signal. Clinical, emotional and developmental aspects of infant and toddler crying, MacKeith Press, 2000.)

So in addition to the billions of parents who have been listening more-or-less closely to their infants in the ordinary course of life, there have been hundreds of researchers (thousands, including all their students and assistants) who have examined and tested millions of recorded infant cries, natural and provoked, over decades of work, proposing and debating a wide variety of ideas about the cries' function and information content. It's possible that there's a simple, universal "lexicon" of five cry (or pre-cry) "words" that these billions of parents and thousands of researchers have missed -- but it seems implausible, frankly.

All the same, Dunstan's insistance that those five universal words are really there may not be a bad thing at all. As Rami Nader, Elizabeth A. Job, Melani Badali and Kenneth Craig say in a comment on Soltis' BBS article ("Infant crying in context"):

Our focus has been on the role of early cry as a commanding source of information about infant pain and distress that requires interpretation by an adult caregiver. Its inherent ambiguity may offer an adaptive advantage, as resolution requires adult presence and scrutiny of other behavioral, physical, and contextual factors.

And it makes sense that paying close attention to an infant's vocalizations, and trying to correlate them to other behavioral and contextual factors, can only be a good thing. There's some experimental evidence for this piece of common sense, in B.M. Lester, et al., "Developmental Outcome as a Function of the Goodness of Fit Between the Infant's Cry Characteristics and the Mother's Perception of Her Infant's Cry", Pediatrics 95(4) 516-521, 1995:

Objective. To determine whether the "goodness of fit" between infant cry characteristics and the mother's perception of the cry is related to developmental outcome at 18 months of age.

Design. This was a prospective, longitudinal study from birth to 18 months performed in a blinded manner.

Setting. The study was conducted in a maternity hospital, including normal and special care nurseries and a laboratory for developmental follow-up.

Patients. The 121 term and preterm infants and their mothers were selected to meet medical criteria.

Measurement. Acoustic analysis of 1-month infant cry and the mother's perception of the same cry was used to divide subjects into four groups representing matches and mismatches between infant cry characteristics and maternal cry perception. Primary outcome measures of cognitive, language, motor, and neurologic outcome were administered at 18 months. Caretaking environment measures were also recorded.

Results. Statistically significant (P < .05) findings showed that matched groups scored higher on measures of language and cognitive performance than infants in the mismatch groups, with a particular advantage for infants in the matched group in which mothers accurately perceived the higher-pitched cries of their infants. There were no differences between the groups in biologic or sociodemographic factors. Group differences were observed in social support and maternal self-esteem.

Conclusions. Matches and mismatches between infant cry characteristics at 1 month and the mother's perception of the cry are related to cognitive and language outcome at 18 months in term and preterm infants. This relation is probably due to transactional processes in which developmental outcome is affected by the clarity of the infants' signals and by the ability of the mother to accurately perceive her infant's signals. The mother's ability to read her infant's cues may be affected by factors such as social support and self-esteem.

[For a sample of previous analysis of the acoustic features of infant cries, and their perceptual analysis, see e.g. Athanassios Protapapas, "An assessment of the perceptual role of individual acoustic features of infant cries", Brown University M.S. thesis, 1993, or Athanassios Protopapas and Peter Eimas, "Perceptual differences in infant cries revealed by modification of acoustic features", J. Acoust. Soc. Am. 102(6) 1997.

Previous studies of infant cry acoustics and their perceptual significance have remained inconclusive as to the graded nature of cry production and perception and to the exact role and importance of particular acoustic features. In this study, a set of infant cries were digitally analyzed and resynthesized to form natural-sounding cries with varying fundamental frequency (F0), degrees of jitter (period to period variations in F0), and rise time (time for F0 to reach its maximum value). In a perceptual rating task, higher-F0 cries as well as cries with larger amounts of jitter tended to be given more negative ratings than were lower-F0 cries and cries with less jitter, respectively. The perceptual ratings of the rise time manipulations were inconclusive. This study demonstrated a perceptual effect of F0 and jitter independently of other parameters, consistent with current notions of infant cry gradedness. It was also shown that digital signal processing techniques can be fruitfully applied to infant cry research.


[And Paul J. Camp points out that The Simpsons -- of course -- have already explored this area of science (in episode 8F23, Brother, Can You Spare Two Dimes?):

Herb stands in a public park trying to figure out an idea, when a woman struggles to understand what her baby's trying to tell her.

Herb: [confronting her] Lady, you just gave me the idea of a lifetime! How do I thank you?
Lady: Please don't hurt me.
Herb: Consider it done.


Herb invites the family in for a presentation that will change the world, and brings out a drinking bird. Homer is in awe. Herb tells him to take it easy, but Homer continues to ogle the bird. Herb introduces his true plan: a baby translator.

It measures the pitch, the frequency, and the urgency of a baby's cry, and then tells whoever's around, in plain English, exactly what the baby's trying to say! Everything from "Change me" to "Turn off that damn Raffi record!"


He reveals his less-than-spectacular creation. Marge "oooh"s. Herb says "You don't have to humor me", but she retorts that it's pretty ingrained. Homer says it's the stupidest thing he ever saw. [...]

Just then Maggie reveals its ability.

Lavish attention on me, and entertain me.

Everyone is in awe. Lisa plays peekaboo with her.

Lisa: Maggie? Maggie? [covers her eyes]
Maggie: [babbles]
Translator: [monotone] Where did you go?
Lisa: Peekaboo! [uncovers eyes]
Maggie: [laughs]
Translator: [monotone] Oh, there you are. Very amusing.


Posted by Mark Liberman at December 7, 2006 07:36 AM