Infant-directed Speech by Dutch Fathers: Increased Pitch Variability within and across Utterances

ABSTRACT

Although both fathers and mothers speak differently in infant-directed speech (IDS) compared to adult-directed speech (ADS), the acoustic characteristics of present-day paternal IDS are still insufficiently understood. To extend this understanding, 11 fathers and 17 mothers in The Netherlands were recorded interacting with their infant (260–476 days old; for IDS) and with an adult experimenter (for ADS). Both fathers and mothers were found to raise their average pitch, expand their pitch variability within utterance, and increase their pitch variability across utterances in IDS. Moreover, fathers increased their pitch variability within and across utterances more than mothers. The IDS produced by present-day Dutch-speaking fathers is thus acoustically highly dynamic, in line with fathers’ energetic interaction style.

Acknowledgments

This work was funded in part by a Netherlands Organisation of Scientific Research grant awarded to Titia Benders. Thank you to our participants, to Gisela Govaart, Maartje van der Hoeve, Amber Foster, and Shannon Konara for data coding, as well as to Floor Arts for her indirect intellectual contribution through a related project. Thank you to the audiences at the Experimental Psychology Conference held at Melbourne University (2016), the Australian Fatherhood Research Network Symposium at the University of Newcastle (2016), and the International Conference on Infant Studies (2016) for their input to previous versions of the work presented in this manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ F0 in paternal IDS has been found higher than F0 in their IDS if measured as the mean in each utterance (Amano et al., 2006; Fernald et al., 1989; Jacobson et al., 1983; McRoberts & Best, 1997; Van de Weijer, 1997), as the perceived primary stress of each utterance (Papoušek et al., 1987), or as the mode across the entire conversation (Warren-Leubecker & Bohannon, 1984). These detailed measurement considerations have not received much attention in the literature and extend beyond the scope of the present review but will be acknowledged for completeness’ sake using footnotes.

² Note that Shute and Wheldall (1999) observed an increase in F0 during book reading when considering the mode rather than the more typically used mean across samples.

³ Either gender may raise F0 more, depending on whether the mean or mode across the interaction is considered (Shute & Wheldall, 1999).

⁴ Moreover, variability in both domains can be instantiated as the range (maximum minus minimum F0 in the domain) or as the standard deviation (taken over all F0 samples in the domain). This distinction extends beyond the scope of the present study.

⁵ McRoberts and Best (1997) and Van de Weijer (1997) each present a case study of one couple, with at least one of the participating parents being a full-time academic. Rondal (1980) presents a study on IDS by French-speaking parents, but is not further discussed further here, as it contains no measures related to the parents’ pitch.

⁶ As indicated, these data were recorded from parents and infants who visited the lab for a speech perception experiment and concordant recording of parent-infant interaction. A total of 50 dyads have been recorded in this fashion. Of these, the 18 mother-infant dyads who were successfully recorded twice have been reported in Benders (2013). The 17 mother-infant dyads reported in the present study include 6 dyads for whom a second session could not be scheduled, 2 dyads for whom the father attended the second session, 3 dyads for whom a second recording was made but not available due to technical or procedural issues, and 6 dyads who were only invited for a single session for reasons unrelated to the infant’s or mother’s performance.

Some readers might be concerned about 8 out of 17 mother-infant dyads in the present study having experienced some scheduling issue, as this might indicate an oversampling of less motivated mothers. This was not, however, our subjective impression of the participants: Participants had not been asked to commit to two sessions when they agreed to the first session and were typically willing to participate in a second session, but unavailable due to holidays or other standing commitments.

Other readers might wonder about the relatively high proportion of father-infant dyads in the total sample of 50 dyads. Across the 50 recordings reported in Benders (2013), the present study, and the data excluded from both studies, mother-infant dyads accounted for 37 recordings (74%; 18 in Benders (2013) + 17 in the present study + 2 excluded); father-infant dyads accounted for 13 recordings (26%; 11 in the present study + 2 excluded). As not all infant studies conducted between 2009 and 2011 in the University of Amsterdam Phonetics Lab included the IDS recording sessions, these are not exact measures of the proportion of infants accompanied by mothers versus fathers. However, it does reflect the subjective experience testing infant participants in the Phonetics Lab at the University of Amsterdam that a non-negligible proportion of infants came to the lab accompanied by their father, usually on the fathers’ day off work.

⁷ Many thanks to the editor for prompting us to inspect the data for this issue.

⁸ The primary difference between the initial script used for the present project and the simplified To TextGrid (silences) … function is that the latter detects candidate utterances entirely on the basis of the intensity envelope. Informal inspection of the data revealed that the script’s most common errors included disregarding voiceless speech sounds as potential utterances, including utterances spoken in a creaky or whispered voice quality). The To TextGrid (silences) … function is less likely to make such errors and may thus be preferred by most users.

⁹ The following R code illustrates that the standard deviation is larger in a sample of 10k than in a sample of 10 in around 56% of simulations:

nSim = 10000; result = numeric(nSim)

for (i in 1:nSim) {s1 = sd(rnorm(10)); s2 = sd(rnorm(10,000)); result[i] = as.numeric(s2>s1)}

sum(result)/nSim

Additional information

Funding

This work was supported by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek. Grant number 021.002.095 awarded to TB.