Impaired word stress pattern discrimination in very-low-birthweight infants during the first 6 months of life


* Correspondence to first author at Klinik für Neonatologie, Campus Virchow-Klinikum, Charité - Universitätsmedizin Berlin, Augustenburger Platz 1, 13353 Berlin, Germany.


Prosodic information, such as word stress and speech rhythm, is important in language acquisition, and sensitivity to stress patterns is present from birth onwards. Exposure to prosodic properties of the native language occurs prenatally. Preterm birth and an associated lack of exposure to prosodic information are suspected to affect language acquisition in preterm infants. Fifty healthy very low birthweight (<1500g) preterm German infants (24 males, 26 females; mean gestational age [GA] 27.6wks, range 26.4–29.9) and 103 comparison term infants (48 males, 55 females; mean GA 40wks, range 39.4–40.8) were recruited. Prosodic discrimination performance was assessed using the head-turn preference paradigm, an objective behavioural psycholinguistic test for measuring orientation time (OT) to auditory stress patterns. Among matched preterm and term infants, preterm infants (n=30) did not differentiate stress patterns at the corrected age of 4 or 6 months. In term infants (n=30), the OT was longer towards the trochaic (stress on first syllable, characteristic for German) than the iambic (second syllable) stress patterns (11.64 vs 9.18s, p<0.001, and 11.02 vs 8.32s, p<0.001, at 4 and 6mo respectively). Neurodevelopmental scores (Bayley Scales of Infant Development, 2nd edn) were not different from reference values in both groups of infants. Preterm birth and deficient early prosodic information affect prosodic processing during the first half year of life.

The acquisition of spoken language requires the identification of lexical elements, syntactic relations, and regularities. To this end, an infant has to segment continuous speech input into separate units such as sentences, clauses, and words. One of the earliest cues that an infant may use for discovering separate units is prosody, namely word stress, sentence intonation, and speech rhythm.1 Word stress is defined as a prominence relation between the syllables (and metrical feet) of a phonological word, with prominence characterized by length, intensity, and pitch accent.2 From birth onwards, infants are sensitive to different kinds of prosodic information. They respond to prosodically marked boundaries in the signal1 and they discriminate stressed from unstressed syllables3 as well as between different languages by means of rhythmic differences.4

In English and German about 90% of all bisyllabics are trochaic words.5,6 The trochee is a strong–weak accentual pattern with stress on the first syllable (máma) compared with the iamb with stress on the second syllable (mamá). At the age of 7 to 8 months, English infants use this prosodic cue for word segmentation.7 At the age of about 9 months, English infants prefer a trochaic over an iambic stress pattern.8 German infants have been shown to prefer a trochaic stress pattern at the age of 6 months, according to behavioural response studies.9 Very recent studies in 4-month-old German and French infants elicited electrophysiological responses that showed a processing advantage for the native stress pattern.10

Infants’ capabilities for the fast acquisition of the native language’s prosodic features may be connected with prenatal exposure to prosodic information during pregnancy by external acoustic stimulation. Although the low-pitched speech signal in utero is unsuitable for sound identification, rhythmic and intonational as well as different stress patterns are easily heard.11 The auditory system of the human fetus is functionally developed as early as during the 23rd to 25th week of gestation,12 and several studies have demonstrated physiological responses by external acoustic stimulation.13 Whereas a term infant experiences prosodic information up to birth in the form of low-passed filtered sound, very low birthweight (VLBW) infants are exposed to the non-physiological high-frequency ambience of a neonatal intensive care unit in a closed incubator right at the time when hearing begins.

We hypothesized that less exposure to native language prosody associated with VLBW may result in a detectable deficit or delay in infants’ processing of prosodic information. To test this, healthy, normally developing term infants were compared with normally developing VLBW pre-terms (<1500g, corresponding to <32wks’ gestation in children with weight appropriate for gestational age14). The study focused on whether preterm birth is associated with specific differences in response to prosodic features of the ambient language during the first half year of life.


Recruitment and participant flow

From December 2004 until June 2006, consecutive singleton neonates at two sites of one university hospital (Charité - Universitätsmedizin Berlin, Germany) were screened for participation in this longitudinal study. Parents of the infants expressed their consent by returning the consent form by surface mail within a few days of being informed of this study. Approval by the ethics committee at Potsdam University (Germany) was obtained in 2003. Infants were eligible for inclusion in the ‘preterm’ study group if their birthweight was less than 1500g (VLBW infants).

Throughout this report, the preterm infants’ postnatal age is given as corrected age: i.e. as age after the expected date of delivery. Comparison infants were eligible for inclusion in the ‘term’ infants group if they were born at term (between 37 and 42wks’ gestation) after an uncomplicated pregnancy and delivery. Furthermore, eligible newborns were required to be of appropriate weight for gestational age at birth and to be born into monolingual German families.

Infants were recruited for participation in this study by the attending neonatologist provided that: (1) physical examination did not reveal manifest developmental problems; (2) standardized hearing testing using otoacoustic emissions was not abnormal; and there was no history of (3) acidosis or asphyxia, (4) intraventricular haemorrhage (grade I to IV according to Papile15) or periventricular leucomalacia, (5) bronchopulmonary dysplasia, (6) hydrocephalus, (7) chromosomal abnormality, (8) metabolic or syndromal disease, or (9) malformation or other major medical problem. Hyperbilirubinaemia treated solely with phototherapy, non-complicated infections with interleukin-6 levels less than 30pg/ml, and necrotizing enterocolitis (Walsh stage no more severe than IIa)16 did not result in exclusion from this study.

Preterm infants were nursed in closed double-wall incubators (type 8000 IC, Draegerwerke Lübeck, Germany; average internal sound level, 45dB) until they reached a weight of about 1500g. The neonatal intensive care unit adhered to recommended standards of care for neonates in a quiet environment.17 Preterm infants were regularly examined by a specially trained neonatologist for physical and neurodevelopmental development, including evaluation of the ‘early motor pattern profile’.18 Term infants were examined by their respective paediatricians at nationally recommended time points, the documentation of which was checked. Preterm and term infants were assessed with the Bayley Scales of Infant Development, 2nd edn, at the (corrected) age of 6 months.19

Baseline characteristics of the recruited infants are given in Table I. No infants were additionally recruited for testing at 6 months. The progress of creation of the final analysis group is shown in Table SI (supplementary material available online). There was no difference in baseline characteristics between those preterm and term infants who were or were not included in the final analysis group with respect to sex distribution, birthweight, gestation duration, umbilical artery pH, and Apgar value at 5 minutes, as well as (in preterm infants) duration of hospitalization, ventilatory support, and of incubator care.

Table I.   Baseline and clinical characteristics of all recruited preterm and term infants
Characteristic/ groupPreterm (n=50)Term (n=103)pb
  1. Values are given as median (interquartile range) or n (%). aIncluding intermittent positive pressure ventilation and continuous positive airway pressure. bp values according to two-sided χ2 test (sex) or two-sided t-test for unpaired samples (other variables) for the comparison of preterm versus term infants.

Gestational age, wks27.6 (26.4–29.9)40 (39.4–40.8)<0.001
Birthweight, g1012 (738–1295)3600 (3250–3910)<0.001
Sex, Male/Female24/26 (48/52)48/55 (47/53)0.975
Umbilical artery pH7.295 (7.26–7.33)7.27 (7.24–7.31)0.359
Apgar score at 5 min7 (6–8)10 (10–10)<0.001
Ventilator supporta41 (82)0 
Supporta duration, d15 (3.8–32) 0 
Incubator care, d41 (27.3–63.5)0 
Hospitalization, d63 (43.5–88)0 

At the scheduled testing times (4 and 6 months of age), the actual age was (median [interquartile range]) 123 (121–126) days and 182 (178–185) days in preterm infants, and 122 (120–125) days and 182 (179–184) days in term infants respectively.

Experimental settings

Behavioural experiments were performed at 4 and 6 months of age under the standardized head-turn preference paradigm (HTP) first published by Hirsh-Pasek et al.20 These experiments recorded the length of the orientation time (OT) during which the child turned its head towards varying auditory stimuli and continued to listen to them. The length of the OT indicates infants’ preferences as well as their ability to recognize a difference between two stimuli. HTP OT measurements have been shown to be associated with later language outcome.21

Each experiment lasted between 3 and 5 minutes, depending on the child’s compliance. The HTP experiments were performed with New Emitter Start Unit (NESU) software (version 3.1, Max-Planck-Institute for Psycholinguistics, Nijmegen, the Netherlands), a multimedia software for stimulus presentation.

Two consonant–vowel sequences, either with trochaic or with iambic stress pattern, served as stimuli. The stimuli were recorded by a female German native speaker. From these recordings five trochaic and five iambic speech files (gába vs gabá) were created. Each speech file contained different tokens of the same rhythmic pattern with an interstimulus interval of 600ms. The trochaic speech files contained 16 tokens (mean duration 18.39s). The iambic speech files contained 15 tokens (mean duration 18.01s). To control for order effects, the presentation of the speech files during the experiment alternated between trochaic and iambic speech files with the type of first stimulus varying across participants. This allowed for comparisons of OTs for a pair of one trochaic and one iambic file, taking into account the position of this pair in the course of the experiment. The very first trochaic and iambic stimulus pair served as warming-up trial and was not included in the analysis.

In experiments conducted at 4 months of age only, the infant was familiarized with the trochaic speech file presented for 30 seconds on each side. In this experimental setting, the recognition of, and hence the ability to discriminate between, the different stress patterns were tested. At 6 months of age the infant was not familiarized, to test the spontaneous preference for the native trochaic stress pattern.

During the experiment, the parent was seated on a chair in a test booth holding the infant sitting on their lap (Fig. 1). A green light was placed in front, and red lights were placed to the sides or diagonally at the front, to accommodate the limited head mobility of 4-month-old preterm and term infants (Fig. 1a). Loudspeakers were hidden behind the red lights. The researcher observed the child on video. To catch the infant’s attention the experiment began by flashing the green light. When the infant focused on it, it was switched off, and the right or left red light started to flash. After the infant turned his head, the auditory stimulus was played on this side. The stimulus presentation ended when the infant turned his head for more than 2 seconds, or when the end of the speech file was reached. The total length of time that the infant had its head turned listening to the stimulus was recorded as the OT, not including periods of inattention or movement in between.

Figure 1.

Experimental setting for the head-turn preference paradigm (HTP) for use with infants at (a) 4 and (b) 6 months of age.

Statistical Analysis and Sample Size Calculation

The primary endpoint in this study was the comparison of OTs after trochaic and iambic stimuli in HTP experiments with preterm and term infants at 4 and 6 months of age (corrected age in preterm infants) respectively. Data from HTP experiments were collected in NESU’s result files, which were automatically assembled into a database. For all data processing and statistical analyses, the software R, version 2.6.2, was used.22

Sample sizes were calculated for designing the study based on mean OT per experiment. From OTs in pilot experiments, a meaningful difference between pairs of trochaic and iambic stimuli of 2000ms, an SD of the pairs’ differences of 3500ms, and an overall mean of 10 000ms were assumed. Accordingly, a two-sided t-test at a significance level of 5% would yield a power of about 85% for comparing pairs of trochaic and iambic stimuli in 30 individuals.

The rate of non-analyzable tests was assumed to amount to about 40% in preterm and term infants during the course of the study, for various reasons including behavioural test failures, study withdrawal, development of an exclusion criterion as listed under ‘Recruitment and participant flow’, or acute medical disease at the time scheduled for testing. Accordingly, 50 preterm infants were recruited. For each of these, two term infants (totalling 100 term infants) were recruited as potential matches for a given preterm infant (see below), because the rate assumed above would correspond to the likelihood that there would be only one term infant remaining for matching.

For analysis, two groups were defined to include pairwise matched preterm and term infants. First, the group of 30 preterm infants was defined by having completed the experimental tests and not having developed an exclusion criterion as mentioned above. Subsequently, a term infant fulfilling the same criteria was matched to a given preterm infant according to sex and first stimulus presented, and as closely as possible with regard to date of birth and age at testing. Results are given for the analyzed groups of 30 preterm infants and their 30 matched comparison infants.

For statistical comparison of clinical data, a two-sided χ² test and a two-sided t-test for unpaired samples were used. To compare OT measures by stimulus, a two-sided t-test for paired samples was applied using the experiment-wise sum of the OTs.

For the main analysis, repeated measures analyses of variance (ANOVA) was performed. To estimate the variability of this ANOVA, it was subjected to a bootstrap procedure from which 95% confidence intervals of the p values were estimated.23


For all experiments, the mean OTs measured using trochaic stimuli were 2.3 seconds (SD 1.7s) longer than those measured using iambic stimuli, consistent with the hypothesis of preference for the ambient language’s stress pattern.

The distributions of OTs showed large overlap, and potential differences between preterm and term infants were masked by overall variability (Table II, top). For the main statistical analysis of the data, repeated measure ANOVAs were conducted with stimulus category (trochaic vs iambic) and stimulus pair position (1st to 4th trial) as within-participant factors, and birth (preterm vs term), sex, and type of first stimulus (first stimulus trochaic vs first stimulus iambic) as between-participant factors. The main effects and significant interactions of these analyses are reported in Table III. At both 4 and 6 months, interactions of stimulus category with birth were significant (both p<0.001). In the term group, OTs were significantly longer for the trochaic compared with the iambic stimuli, whereas in the preterm group, OTs were not significantly different by stimulus (t-tests, Table II).

Table II.   Results of orientation time (OT) measurements by experimental setting in preterm (n=30) and term (n=30) infants (final analysis group)
Group of infantsAgeaat test, moComparisonMean (SD) OT towards trochaic stimulus, sMean (SD) OT towards iambic stimulus, sp (t value; degrees of freedom)
  1. aCorrected age in preterms infants.

Preterm and term4Premature vs term birth10.630 (5.409)9.227 (4.964)0.2 (−1.2; 58)
Preterm and term6Premature vs term birth10.610 (4.974)9.153 (4.732)0.4 (0.54; 58)
Preterm4Trochaic vs iambic stimulus9.625 (5.653)9.273 (5.206)0.5 (0.66; 29)
Preterm6Trochaic vs iambic stimulus10.200 (5.140)9.929 (4.926)0.4 (0.65; 29)
Term4Trochaic vs iambic stimulus11.640 (4.974)9.182 (4.731)<0.001 (7.1; 29)
Term6Trochaic vs iambic stimulus11.020 (4.789)8.377 (4.416)<0.001 (6.12; 29)
Table III.   Results of multifactorial analyses of repeated orientation time measurements by experimental setting in preterm (n=30) and term (n=30) infants (final analysis group)
FactorLevelsTypedfFpp95% CI
Preterm and term infants at 4moa
 StimulusTrochaic vs iambicu120.8<0.001<0.001–0.020
 BirthPreterm vs termu11.450.230.007–0.905
 SexMale vs femaleu11.010.320.014–0.915
 First stimulusTrochaic vs iambicu10.070.790.057–0.968
 StimulusPair positiono34.770.003<0.001–0.473
 InteractionStimulus with birthu111.720.001<0.001–0.199
 InteractionStimulus with sexu15.130.027<0.001–0.587
 InteractionStimulus with pair positiono30.640.590.016–0.960
Preterm and term infants at 6moa
 StimulusTrochaic vs iambicu123.2<0.001<0.001–0.014
 BirthPreterm vs termu10.290.60.041–0.956
 SexMale vs femaleu10.010.90.060–0.964
 First stimulusTrochaic vs iambicu11.570.220.006–0.892
 StimulusPair positiono311.44<0.001<0.001–0.005
 InteractionStimulus with birthu115.4<0.001<0.001–0.069
 InteractionStimulus with sexu10.570.450.022–0.954
 InteractionStimulus with pair positiono30.420.740.015–0.955
  1. aCorrected age at test in preterm infants. Group, each group corresponds to an experimental setting and to one ANOVA. Factor, all main effects and two-way interactions are listed. Levels, levels, units, or components (in case of interactions) of the factor. Type, factors were unordered (u), ordered categorical (o), or continuous (c). df, degrees of freedom; F value of the factor; p values of <0.05 are significant; p95% CI, 95% confidence interval according to p value bootstrap estimates.

Preterm infants at 4 and 6moa
 StimulusTrochaic vs iambicu10.0060.939
 SexMale vs femaleu10.0020.966
 Age at test, mo4 vs 6au11.050.317
 Ventilatory support, d c10.500.489
 Gestation duration, wks c11.230.280
 Umbilical artery pH c10.360.557
 Apgar score value at 5min c13.980.060
 Incubator care, d c10.210.651
 Hospitalization, d c10.240.630
 Stimulus with pair position o34.740.003

Exploring OT measurements in preterm infants only, no significant association of the OTs was found with stimulus category, age at test (4 or 6mo), baseline or clinical characteristics (Table III, bottom). The Apgar value at 5 minutes had the highest F value (3.98; 1 degree of freedom, p=0.059).

Among the clinical characteristics, the duration of ventilation support (producing extra noise from air flow for continuous positive airway pressure or Synchonous intermittent mandatory ventilation) and the gestation duration (as surrogate for intrauterine exposure to language) were not associated with OT measurements and stimulus preference in preterm infants (Spearman’s rank correlation, p=0.941 and p=0.824 respectively).

The Bayley Scales of Infant Development, 2nd edn, scores of mental development were similar in the preterm (median [interquartile range] 100 [94–110],) and term infants (100 [94–110], p=0.366), as were the scores of motor development in the preterm (91 [77–97]) and term infants (92 [83–100], p=0.282), at 6 months of (corrected) age.


The results of this study indicate that in VLBW infants, prosodic processing is deviant during the first 6 months of life compared with matched term infants. In contrast to the term infants, preterm infants, as a group, did not show evidence of discriminating the trochaic from the iambic speech stimuli at the corrected age of 4 months or of preferring the trochaic stress pattern at 6 months of age, as did the term infants. However, the OTs were overall quite similar across the groups, providing no indication of generally slower processing in this group of preterm infants than in term infants.

This developmental pattern implies that infants born at term have learned the specific prosodic features from their native surrounding at the age of 6 months, as shown by a behavioural method. Our results obtained with the HTP are supported by findings using event-related brain potentials, which revealed a diverse response to different stress patterns as early as 4 months of age. This earlier response is attributed to the fact that event-related brain potentials are a direct reflection of neural processing, capable of revealing very rapid implicit processing of familiar stimuli.24

The HTP results of the term infants at the age of 4 months may be interpreted as showing clear detection of different stress patterns after familiarization with the trochaic disyllabics. However, as infants had been familiarized at 4 months of (corrected) age with a trochaic stress pattern, which is also the typical stress pattern of the ambient language, the experiments could not discriminate the effects of the native language from that of the familiarization, allowing the hypothesis that effects of the native language’s prosody might already be shown in behavioural measures at 4 months.

Nevertheless, at 6 months of age, the longer OT measurements in term infants indicate a preference for the native stress pattern, which is probably not lost during further maturation. Despite similar mean OTs in both groups, we cannot exclude the possibility that an increased interstimulus interval would result in discrimination of different stress pattern in the preterm group as well.25 Similarly, an influence of sex and of lower Apgar values in the 30 preterm infants studied on the observed processing performance cannot be excluded.

Eutrophic VLBW infants, in whom the definition of less than 1500g corresponds to fewer than 32 weeks of gestational age, are less exposed to intrauterine stimuli than term infants by 1 to about 3 months. In addition, medical care necessitates closed-wall incubator nursing. So in this study, preterm infants were practically not exposed to prosodic information for a further 41 days (Table I), or until almost 6 weeks of postnatal age. Furthermore, particularly in quiet neonatal intensive care units, parents do not exaggerate speech prosody as much as in noisier units.26

About 20 to 40% of all surviving VLBW infants were found to be impaired in language development at preschool or school age.27 Several follow-up studies of preterm infants reported on specific difficulties in language acquisition, such as deficiencies in language comprehension and auditory discrimination,28 lower scores for vocabulary, expressive language, phonological short-term memory, and general non-verbal ability,29,30 and reduced syntactic performance.31 Furthermore, poor skills in writing and reading were obvious at school age.32 Others did not find significant impairment in language development or test performance in preterm infants.33,34 Our results suggest that first indications of language peculiarities may already be found in the early processing of prosodic information by preterm infants.

It is unclear whether the ability to discriminate word stress patterns is only delayed, whether preterm infants simply need more time to learn prosodic features, or whether the prosodic sensitivity is actually impaired, relating to subsequent detriment to lexical development as shown by electrophysiological evidence.35 It is tempting to speculate that preterm infants might use different strategies to acquire language, for instance by putting a heavier weight on statistical distributions of words and syllables within sentences, as assumed in the statistical learning framework.36 Further studies are needed to clarify these questions.


We are grateful to Joachim Dudenhausen, head of the Department of Obstetrics, Campus Virchow-Klinikum, Charité - Universitätsmedizin Berlin, for allowing the recruitment of term infants, and to Christoph Bührer, Ingrid Grimmer, and Jürgen Weissenborn for scientific consultancy. We especially thank Cindy Gatzke for investigating the developmental and neurological state of the infants, Christian Peiser for recruiting the infants and obtaining informed consent from their guardians, Angela Haesner for conducting hearing screening examinations, and Werner Hopfenmüller for statistical expertize. This study was supported by the Deutsche Forschungsgemeinschaft (German Research Foundation, DFG), grant numbers HO1960/10-2 and AOBJ: 521782.