Affect in Infant-Directed Speech of Swedish-Speaking Mothers and Fathers to 3-, 6-, 9-, and 12-Month-Old Infants

ABSTRACT

Infant-directed speech (IDS) is characterized by a range of register-typical characteristics. Many of those can be objectively measured, such as acoustic-prosodic and structural-linguistic modifications. Perceived vocal affect, however, is a socio-emotional IDS characteristic and is subjectively assessed. Vocal affect goes beyond acoustic-prosodic and structural-linguistic IDS features and includes a perceptive-subjective component in the listener. This study describes vocal affect valence in Swedish IDS during the first year of life, and compares vocal affect between mothers’ and fathers’ IDS and their adult-directed speech. Adult native speakers of Swedish (N = 16) rated affect valence in low-pass filtered IDS samples from free play interactions of mothers with infants at 3, 6, 9, and 12 months (N = 10) and fathers with their 12-month-olds (N = 6). Across the first year, the trajectory of mothers’ positive affect in Swedish IDS shows the highest affect scores toward the youngest infants and the lowest at 9 months of age. This follows a pattern comparable to that in Australian English IDS, showing that mothers express vocal affect similarly across different languages and cultures. Both mothers’ and fathers’ IDS to 12-month-olds have higher positive vocal affect valence than their adult-directed speech. There was no difference in positive affect valence between mothers’ and fathers’ IDS, that is, mothers and fathers express vocal affect to the same extent when talking to their infants. In conclusion, the findings of this study indicate that high positive vocal affect characterizes IDS across different languages and speakers.

Introduction

Infant-directed speech (IDS) is the speech register typically used when communicating with infants, regardless of language (e.g., Fernald & Simon, 1984; Kitamura et al., 2002; Papoušek et al., 1991; Werker et al., 1994). It differs from adult-directed speech (ADS) both in acoustic-prosodic and structural-linguistic regards. Acoustic-prosodic modifications in IDS are – in comparison to ADS – higher fundamental frequency (f_o),¹ wider f_o excursions, more distinctive f_o contours (e.g., Fernald & Mazzie, 1991; Fernald & Simon, 1984; Kitamura & Lam, 2009), steeper spectral tilt (Shinya et al., 2009; Wang et al., 2016), slower speech rate (Cristià, 2010; Fernald & Simon, 1984), shorter utterance and pause duration (U. Marklund et al., 2015), more pronounced syllable-lengthening (Albin & Echols, 1996; Martin et al., 2016), and more hyperarticulated vowels (Burnham et al., 2002; Kalashnikova & Burnham, 2018; Kuhl et al., 1997; E. Marklund & Gustavsson, 2020; Odijk & Gillis, 2022). Structural-linguistic modifications include shorter utterance length, utterance-final positioning of target words (Aslin et al., 1996; Fernald et al., 1989) and more repetition than in ADS (Bortfeld & Morgan, 1999; Fernald & Morikawa, 1993). Some IDS modifications remain stable across development, for example, f_o variability and degree of hyperarticulation, while others vary with infant age, for example, mean f_o, articulation rate, and vowel duration (Cox et al., 2022). All these modifications can be measured objectively when analyzing speech recordings and transcriptions. However, IDS also exaggerates positive vocal affect, a modification that cannot be directly accessed from the speech signal as it manifests across several prosodic features and is quantified by subjective listener ratings.

Relationship between prosody and vocal affect

Vocal affect manifests in through prosodic means, although the two components are distinct. Prosody can be directly accessed in the speech signal and objectively measured, whereas vocal affect requires subjective ratings for quantification, even though it can be approximated by a number of acoustic-prosodic characteristics. Positive vocal affect is predominantly expressed by higher f_o, wider f_o range and higher intensity (Johnstone & Scherer, 2000), and associated with higher frequencies across all formants, likely due to the articulatory impact of smiling on the acoustic signal (Benders, 2013; Dodane & Al-Tamimi, 2007; Englund & Behne, 2005; Green et al., 2010). The predominant perceptual cue to positive vocal affect is exaggerated prosody (Banse & Scherer, 1996).

It has been suggested that some of the acoustic-prosodic modifications in IDS relative to ADS are in fact a result of the high amount of positive affect in IDS, for example, that observed hyperarticulation could be a consequence of a shortened vocal tract when smiling (Benders, 2013). In line with this, only minor differences in mean f_o and rhythmic tempo remained between IDS and ADS when highly affective ADS was elicited with the help of imaginary scenarios in fixed carrier sentences (Trainor et al., 2000).

Positive vocal affect in IDS

IDS conveys more positive vocal affect than ADS (Kalashnikova & Burnham, 2018; Katz et al., 1996; Kitamura & Burnham, 2003). A study on Australian English IDS reports an inverted S-shaped pattern of positive vocal affect across the first year of life: In IDS to newborns, affect valence is relatively low (although higher than for ADS), increases to peak in IDS at around 6 months, and then declines at 9 months with a slight recovery at 12 months (Kitamura & Burnham, 2003). Mothers show different trajectories of vocal affect depending on the sex of the infant. Specifically, positive vocal affect increases with age more in speech to girls than to boys between birth and 12 months (Kitamura & Burnham, 2003).

There are differences between languages in how and to what extent specific IDS characteristics are realized (e.g., Fernald et al., 1989; Takada, 2005). This is likely the case for affect, as the expression of vocal affect via prosodic means is restricted in tone languages, where f_o modulations are used to convey lexical distinctions (Kitamura et al., 2002; Ross et al., 1986). However, to the best of our knowledge, vocal affect in tone language IDS has as of yet not been studied. In Japanese, a pitch-accent language, where lexical information is conveyed by f_o modulations in a limited number of cases (Kubozono, 2012), affect in IDS has only been studied in relation to linguistic content, not prosody (Fernald & Morikawa, 1993; Toda et al., 1990).

Functions of IDS prosody and vocal affect

The functions of IDS are threefold: i) it arouses infant attention (Werker & McLeod, 1989), ii) facilitates language learning (Kalashnikova & Burnham, 2018; Kalashnikova et al., 2017) and iii) creates socio-emotional bonds (Golinkoff et al., 2015; Takada, 2005). For the present study, the first two of these functions are of interest.

IDS captures and maintains infant attention. From birth, infants display a behavioral preference for IDS over ADS (Frank et al., 2020; Dunst et al., 2012; Newman & Hussain, 2006; Panneton Cooper & Aslin, 1990; Panneton Cooper et al., 1997; Pegg et al., 1992). This well-established preference may be driven by the high positive affect typical for IDS rather than the dynamic modulation of f_o. Infants prefer speech with high positive affect over speech with low positive affect when f_o modifications were controlled for (Kitamura & Burnham, 1998), and IDS preference is not present when it is contrasted with high affect ADS (Singh, Morgan, et al., 2002). Infants like happy speech so much that they will prefer happy ADS to neutral IDS, even though IDS contains the higher f_o (Singh, Morgan, et al., 2002). A preference for positive vocal affect has been shown in infants throughout the first year of life (Panneton et al., 2006; Singh et al., 2004; Singh, Morgan, et al., 2002).

IDS has a facilitative effect on a range of language learning tasks, for example, word-learning (Foursha-Stevenson et al., 2017; Graf Estes & Hurley, 2013; Ma et al., 2011), word-recognition (Singh et al., 2009; Singh, Bortfeld, et al., 2002; Song et al., 2010) and statistical learning (Bosseler et al., 2016; Thiessen et al., 2005). Prosodic IDS modifications in particular play an essential role in infant language development, as they have a positive impact on speech segmentation (Thiessen et al., 2005), finding stressed words in sentences (Fernald & Mazzie, 1991), discriminating multi-syllabic words (Karzon, 1985), and parsing clauses (Kemler Nelson et al., 1989).² No association between positive vocal affect and language outcomes has been shown in the few studies on the subject. Specifically, no relationship was found between vocal affect in IDS at 7, 9, 11, 15, 19 months and vocabulary at 15 and 19 months (Kalashnikova & Burnham, 2018), and no impact of positive vocal affect was found on word-learning in 14-month-olds (Schwarz et al., 2019; Schwarz et al., in preparation).

The present study

While no study so far has linked vocal affect in IDS to language outcomes, there are reasons to believe that it nevertheless contributes to language learning. Specifically, positive vocal affect in IDS ensures that infants find speech input interesting (Panneton et al., 2006; Singh et al., 2004; Singh, Morgan, et al., 2002). It is likely that this attention-catching trait of IDS underlies its positive impact on language learning (Nencheva & Lew-Williams, 2022; Saint-Georges et al., 2013). In line with this, infants of depressed mothers – who exhibit less positive vocal affect in their IDS than non-depressed mothers (Lam-Cassettari & Kohlhoff., 2020) – do not learn associations as readily from IDS produced by females, but they do learn from male speakers (Kaplan et al., 2004, 2010). More positive vocal affect on the other hand is produced by mothers when talking to their infants in a simulated hearing loss condition compared to a normal hearing situation, possibly trying to keep infants attentive during interaction (Lopez & Lam-Cassettari, 2018). However, despite its potential relevance for language learning, vocal affect in IDS has been only sparsely studied (Kalashnikova & Burnham, 2018; Katz et al., 1996; Kitamura & Burnham, 2003). Therefore, studying vocal affect in IDS in languages other than English is highly warranted, both descriptively and in relation to language development.

Vocal affect in IDS has so far only been studied in mothers. When it comes to prosodic modifications in IDS, fathers display the same patterns of increased f_o and greater f_o variability in IDS compared to ADS as do mothers (Fernald et al., 1989; Gergely et al., 2017; Niwano & Sugai, 2003; Papoušek et al., 1987). However, the results are not consistent across all languages (Benders et al., 2021; Broesch & Bryant, 2017; Shute & Wheldall, 1999). To access the full picture on vocal affect in IDS, it needs to be studied in both mothers and fathers.

In the present study, perceived vocal affect is mapped out in Swedish IDS, spoken by mothers to 3-, 6-, 9- and 12-month-old and by fathers to 12-month-old boys and girls. We use the established practice of asking naïve listeners to rate vocal affect in low-pass filtered recordings of speech (Burnham et al., 2002; Kalashnikova & Burnham, 2018; Kitamura & Burnham, 2003). Vocal affect in Swedish IDS is of particular interest for two reasons: i) in Sweden, parental leave is relatively equally divided between mothers and fathers (Almqvist & Duvander, 2014), and it is therefore both possible and relevant to study Swedish fathers’ IDS as well as mothers’; and ii) Swedish is a pitch-accent language, which means that f_o occasionally carries lexical information, and vocal affect has not yet been studied in a language where lexical information is conveyed by prosodic modification.

Method

Participants

Participants were Swedish-speaking mothers and fathers who took part with their children in a longitudinal study at Stockholm Babylab, in which play-sessions were recorded (MINT-Project; MAW 2011.0070; Gerholm et al., 2015). Included in the current study were 10 mother-infant dyads (7 boys, 3 girls) recorded when the infants were 3, 6, 9, and 12 months, and 6 father-infant dyads (2 boys, 4 girls) recorded at 12 months.³ Criteria for inclusion was that recordings existed for the desired ages and for an ADS session consisting of a conversation with the experimenter. All families spoke Swedish at home. Parents had a minimum of 12 years of education, and the majority had a university degree. All infants were born full-time and healthy. The families, living in Stockholm County, had originally responded to an invitation to participate in the study, sent to addresses obtained from the Swedish Tax Agency. The study was approved by the Regional Ethics Review Board in Stockholm (2015/63–31).

Speech material

Recording of the parent-infant interaction sessions took place at Stockholm Babylab in a sound-attenuated room with three wall-mounted cameras (Canon HDMI XA10). In addition, the parent wore a Go Pro Hero 3 camera strapped across the chest. Parent and infant were each fitted with a Sennheiser eW 100 G2 microphone. A condenser microphone (AKG SE 300 B) in the room served as backup and was used to synchronize the recordings. The parent was instructed to play freely with the infant on the floor. Besides the free play part of the session from which the IDS recordings used in this study originated, the parent was also asked to perform interactions such as clapping and picture book reading, but these parts were excluded as stimulus material. The play session lasted about 10–15 min.

We followed Kitamura and Burnham’s method to condense the speech material to be rated for vocal affect (Kitamura & Burnham, 2003). Three-minute long audio clips were cut out at approximately 1 minute into each recording, respecting utterance boundaries. Recordings were edited to 25 s clips for ratings and acoustic analyses, by reducing pauses longer than 600 ms to 300 ms, extending shorter pauses to 300 ms, and leaving pauses between 300 ms and 600 ms as they were. Utterances with overlaps of parent speech and infant vocalization were excluded, as were instances of laughter, singing, coughing, and non-speech sounds such as toy noise. To remove semantic content, all IDS and ADS clips were low-pass filtered at 400 Hz. Sound files were then normalized for intensity. Acoustic manipulations were performed in Praat 6.0.21 and 6.0.28 (Boersma & Weenink, 2010). This resulted in a total of 62 clips (40 IDS and 10 ADS samples from mothers; 6 IDS and 6 ADS samples from fathers) to be used as stimuli in the rating experiment.

Prosodic analyses of the stimuli clips were performed on utterance level within each clip after intensity normalization and low-pass filtering. Utterances were defined as speech segments in the clips, separated by a pause of at least 300 ms. The total number of utterances in the clips was 1046 (845 from mothers and 201 from fathers). We included f_o mean, f_o median, f_o minimum, f_o maximum and f_o range (difference between max and min) in the prosodic analyses (see Table 1), all measured in Hz using Praat 6.0.40 (Boersma & Weenink, 2010). We chose f_o median in addition to f_o mean as this measure is less prone to be affected by pitch tracking errors which are more likely to occur when analyzing IDS compared to ADS (see Gergely et al., 2017). To further minimize potential pitch tracking errors, a two-step pitch extraction was carried out (De Looze & Hirst, 2010). In a first step, pitch extraction (autocorrelation) was run with the Praat default settings (floor set to 60 Hz, ceiling to 600 Hz) to create a pitch object. In the second step, the pitch object was used to define the pitch floor and the pitch ceiling in another pitch extraction. Pitch floor and pitch ceiling were set to q15 × 0.83 and q65 × 1.92 with q15 and q65 being the 15^th and 65^th percentile of the original pitch object.

Table 1. The table lists the prosodic measures f_o mean, f_o median, f_o min, f_o max, and f_o range (in Hz) for mothers’ ADS and IDS across the first year of life and in fathers’ ADS and IDS to 12-month-olds. Calculations were performed on utterance level.

Speech style/Age	Speaker	No. utterances mean (total)		fo mean (SD)		fo median (SD)		fo min (SD)		fo max (SD)		fo range (SD)
		Boy	Girl	Boy	Girl	Boy	Girl	Boy	Girl	Boy	Girl	Boy	Girl
IDS 03	Mother	17(121)	16(49)	243(21)	289(21)	233(23)	284(22)	139(13)	164(19)	441(24)	462(18)	302(32)	298(3)
IDS 06	Mother	18(127)	17(50)	240(23)	271(18)	233(21)	263(18)	128(34)	147(28)	443(31)	467(6)	315(20)	321(32)
IDS 09	Mother	16(109)	20(60)	225(27)	250(13)	219(30)	241(16)	139(25)	140(28)	418(12)	459(6)	279(26)	318(28)
IDS 12	Mother	20(138)	21(64)	240(34)	249(30)	232(35)	240(34)	134(17)	147(15)	433(34)	443(14)	298(30)	296(14)
	Father	22(43)	19(75)	165(53)	157(36)	159(56)	150(33)	78(29)	89(18)	378(36)	320(87)	299(7)	231(71)
ADS	Mother	12(83)	15(44)	199(20)	222(32)	197(18)	226(26)	97(32)	100(35)	371(19)	408(36)	274(33)	308(37)
	Father	14(43)	14(68)	121(6)	138(19)	118(5)	135(20)	71(6)	88(11)	245(11)	258(49)	174(6)	170(44)

Rating experiment

Sixteen raters (8 female, mean age = 32 years, SD = 8.9, age range 19–55 years) assessed the vocal affect in the recordings. Raters were recruited via an online recruitment site for research participation. All raters were Swedish native speakers, reported normal hearing, and had at least 9 years of formal education (mandatory basic education in Sweden). Nine participants had university or college level education. For their participation, raters received vouchers for an online book store. An additional 12 raters had to be excluded for either not having Swedish as a native language, not completing the rating study and/or not attending to the stimuli as instructed.

The rating experiment was conducted remotely. Raters were asked to sit in a quiet place with a stable internet connection and to use headphones. Prior to the experiment, participants gave informed consent and provided background information. They were informed about the low-pass filtered nature of the stimuli and instructed to listen to the entire stimulus clip before rating it.

The stimuli clips were presented in random order using online questionnaire software (LimeSurvey 2.05; Schmitz, 2015). Each trial consisted of one stimuli clip being presented, and raters were asked to estimate the affect valence and communicative intent of the speech in that clip. Only affect valence is reported in the present study. Affect valence was rated on a nine-point Likert-scale ranging from −4 to + 4 (Kitamura & Burnham, 2003), with + 4 meaning perceived high positive affect and −4 perceived high negative affect. Raters controlled the pace of the experiment and could listen to each clip as many times as they wanted before moving on to the next. The time spent on each stimuli clip was checked after completion of the experiment and raters who moved on to the next trial faster than the 25 s stimulus duration were excluded from the results.

Results

Statistical analyses were performed in R 4.0.2 (R Core Team, 2021) and the package effectsize (Lakens, 2013).

Correlations between affect valence and prosodic measures

To show the relationship between subjective affect ratings and objectively measured prosodic parameters, correlational analyses were conducted. Pearson product-moment correlation coefficients between affect ratings and the prosodic measures f_o mean, f_o median, f_o min, f_o max, and f_o range were all significant and moderate to strong (α =.05, Table 2, Figure 1).

Figure 1. Significant correlations of moderate size between affect valence ratings and the prosodic measures f_o median, f_o min, f_o max and f_o range. The f_o median is chosen to represent overall prosody in this illustration, rather than f_o mean, as it is less prone to pitch-tracking errors.

Table 2. Significance test results for the correlations between affect valence ratings and prosodic measures.

Prosodic measure	Pearson coefficient	t-value	p-value
fo mean	r = .58	t(12) = 2.513	p = .027
fo median	r = .55	t(12) = 2.265	p = .043
fo min	r = .65	t(12) = 2.973	p = .012
fo max	r = .63	t(12) = 2.835	p = .015
fo range	r = .54	t(12) = 2.207	p = .048

Affect valence in mothers’ IDS

Preliminary analyses of mothers’ affect valence in IDS towards their infants at 3, 6, 9 and 12 months showed only a main effect of infant age, no effect of infant sex and no interaction between infant age and infant sex.

To investigate the effect of infant age on affect valence, we looked at mothers’ speech to infants at 3, 6, 9 and 12 months of age in a within-subjects design. Alpha was set to .05. A one-way repeated-measures ANOVA with infant age as independent variable and affect score as dependent variable revealed a significant difference for infant age with a medium-sized effect size (F_{(3, 13)} = 6.09; p = .001; η² = .07; see Figure 2, left panel). Affect valence rating means for mothers’ IDS were then tested for linear, quadratic, and cubic contrasts over infant age. Vocal affect is rated the most positive for 3-month-olds, decreasing to 6-month-olds and further decreasing to 9-month-olds, while rising again for 12-month-olds as the significant quadratic trend shows (F_(1,10) = 5.301, p = .023). The linear and cubic trends across all four ages were not significant.

Figure 2. Left: Mean affect valence ratings for mothers’ IDS addressing infants at 3, 6, 9 and 12 months of age (box plots depict the mid quartiles of the distribution). Right: Mean affect valence ratings for mothers’ (blue) and fathers’ (red) IDS towards 12-month-olds and their ADS (box plots depict the mid quartiles of the distribution).

Affect valence in mothers’ and fathers’ IDS

To investigate the difference in affect between mothers’ and father’s IDS and ADS, a 2 × 2 ANOVA was conducted with speaker (mother vs. father) and speech register (IDS to 12-month-olds vs. ADS) as within-subject variables. Due to the small sample size, infant sex was not included as a variable. The test showed a significant effect of speech register with a large effect size (F_{(1, 12)} = 54.4; p < .001; η² = .34). No main effect was found for speaker (F_{(1, 12)} = .319; p = .58) nor was the interaction between speech register and speaker significant (F_{(1, 12)} = 3.736; p = .072; see Figure 2, right panel). Both mothers’ and fathers’ IDS is clearly distinctive from their ADS, and mothers and fathers do not differ in degree of vocal affect valence in speech to their 12-month-old infants.

Discussion

The present study shows that perceived vocal affect correlates positively with the prosodic measures f_o mean, f_o median, f_o min, f_o max, and f_o range. This is in line with previous findings (Johnstone & Scherer, 2000), and illustrates the close relationship between subjectively perceived affect and objectively measured acoustic parameters.

The trajectory of vocal affect valence in Swedish mothers’ IDS to 3- to 12-month-old infants follows a quadratic trend, showing a decrease from 3 to 9 months while remaining on the positive part of the scale, increasing again at 12 months. This is similar but not identical to the developmental pattern of vocal affect in IDS in Australian English (Kitamura & Burnham, 2003). The score average of 2 on the Swedish affect valence scale for IDS toward 3-month-olds compares well to the average score of 1.8 for Australian English IDS toward 3-month-olds in the graph of the original Australian study (Kitamura & Burnham, 2003). However, in that study, Australian English IDS has the highest positive affect valence towards 6-month-olds, whereas the Swedish scores decrease linearly from 3 months onwards. IDS in both languages show a decrease in affect valence towards a dip at 9 months (average affect scores: Australian English 1.5; Swedish 1.1) and a recovery at 12 months (average affect scores: Australian English 2.2; Swedish 1.7). Kalashnikova and Burnham (2018) plotted the developmental trajectory of vocal affect IDS to infants from 7 to 19 months, finding the usual high positive affect in IDS compared to ADS, but no effect of infant age. However, their vocal affect scores were based on an average across several different affect types rather than on affect valence only (Kalashnikova & Burnham, 2018) and are thus not directly comparable to the findings of the present study.

The present study is the first to report that vocal affect in fathers’ IDS is more positive than in their ADS, following the same pattern as for mothers. Despite the small sample size, and the fact that only one age group was included, this study gives a first indication of that vocal affect in fathers’ IDS is equal to affect in mothers’ IDS. Given the increased presence of fathers in infants’ early life and the importance of speech input for infant language development (Ferjan Ramírez, 2022), both vocal affect and prosodic modifications in IDS need to be studied further in fathers.

To conclude, high vocal affect catches infant attention, thereby potentially contributing to the positive impact of IDS on language learning (Nencheva & Lew-Williams, 2022). Despite this, vocal affect in IDS remains understudied. The present study contributes with two points of information on vocal affect in Swedish IDS. It shows that vocal affect valence in mothers’ IDS follows a U-shaped pattern over the first year of infancy, and that mothers’ IDS and fathers’ IDS convey similar degrees of vocal affect valence.

Contributions

Author contributions are specified according to the CRediT system (ANSI/NISO Z39.104–2022). Conceptualization: ICS (equal), EM (equal), CLC (equal); Data curation: EM (lead), LG (supporting); Formal analysis: EM (lead), ICS (supporting), LG (supporting); Funding acquisition: ICS (lead), EM (supporting), LG (supporting), CLC (supporting); Methodology: ICS (equal), EM (equal), CLC (equal); Project administration: EM (lead), ICS (supporting); Resources: ICS (equal), EM (equal), LG (equal), UM (equal); Software: EM (lead), LG (supporting); Supervision: ICS (lead), EM (support); Validation: ICS (equal), EM (equal), LG (equal); Visualization: EM (lead), ICS (supporting), LG (supporting); Writing original draft: ICS (lead), EM (supporting), LG (supporting); Writing review and editing: ICS (lead), EM (lead), LG (lead), UM (supporting), CLC (supporting).

Acknowledgments

The authors thank the MINT project (MAW 2011.0070; PI Tove Gerholm) for providing the parent-infant recordings and the MINT families for their participation. We thank the MARCS Institute for Brain, Behaviour and Development and specifically Christine Kitamura for her input. Thanks also to Klara Hjerpe for stimuli creation and data collection.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was funded by the Marcus and Amalia Wallenberg Foundation (MAW 2013.0056, PI Iris-Corinna Schwarz, co-applicant Christa Lam-Cassettari; MAW 2019.0030, PI Ellen Marklund) and Riksbankens Jubileumsfond (RJ P17-0175, PI Lisa Gustavsson).

Notes

¹ Fundamental frequency is abbreviated as f_o, which stands for frequency oscillation, in accordance with abbreviation standards in Phonetics (Titze et al., 2015).

² But see also Spinelli et al. (2017).

³ As many fathers took over parental leave toward the end of the first year and therefore started coming to the play sessions at the lab with their infants, father-infant recordings could only be included in this study at infant age 12 months.