Auditory Word Recognition Ability in Babble Noise and Phonological Development in Children at 3;6 Years of Age

ABSTRACT

Background noise challenges auditory recognition of speech and may reveal the underlying deficits in auditory word recognition skills. Previous studies have reported an association between children’s auditory skills and various linguistic skills, including phonology, although in some languages only. However, language-specific features influence these connections. This study describes the associations between auditory word recognition ability and phonological skills in a representative group of 3;6-year-old children in less studied language, Finnish (N = 65). Auditory recognition was assessed using a closed-set word recognition task presented in multi-talker babble noise (+13 dB) and the phonological skills using Finnish Phonology Test. A significant, moderate correlation was found between Finnish-speaking children’s auditory word recognition and phonological skills. The late-developing phoneme /r/ challenged word recognition. The sibilant /s/ was noted to be the most noise resistant consonant, and the phoneme pair /r/ and /l/ created the most mutual confusion. The accuracy of phonological representations may be a moderating factor for both auditory recognition and phonological skills. Children with strong phonological skills may recognize spoken words more accurately in noisy everyday situations than children with weaker phonological skills. This should be taken into consideration in children’s daily environments, such as daycare centers and preschools.

Introduction

Although it is evident that auditory recognition ability is needed for phonological development, the specific connection between auditory recognition ability and phonological skills in typically developing children is not clear. During auditory recognition, a listener attaches the phonological representation of the word to the lexical representation included in semantic long-term memory (Rönnberg et al., 2010). Phonological development is a continuum from babbling to proto-words and eventually to adult-like speech production. Both auditory recognition ability and phonological skills are developing during childhood years (e.g., Rvachew & Brosseau-Lapré, 2018). In previous studies, the connection between auditory recognition and the development of phonological production has mostly been analyzed in children with speech sound disorders (SSD) (see recent review: Berti et al., 2020; Hearnshaw et al., 2019). Less is known about this relationship in typically developing children. The main aim of the present study is to investigate the association between word recognition ability in background babble noise and phonological skills in a representative group of children aged 3;6 years, an age when phonological skills are still developing.

Background noise and auditory word recognition

When the listening conditions are not optimal, e.g., due to background noise, word recognition becomes challenging (Rönnberg et al., 2010). Further, auditory discrimination deficits may be easier to detect in a noisy environment than in a quiet one (Vance & Martindale, 2012). Children have been noted to be more vulnerable to the harmful effects of noise than adults (Fallon et al., 2000; Klatte et al., 2013). In addition, younger children are hindered more than older children by the addition of background noise (Leibold & Buss, 2013; Nishi et al., 2010; Talarico et al., 2007). Good-enough listening conditions (i.e. children recognizing at least 95% of input) have been shown to vary in relation to the child’s age. For 6-year-olds, the good-enough signal-to-noise ratio is 15.5 dB (signal 15 dB louder than background noise), for 8-year-olds 12.5 dB, and for 11-year-olds 8.5 dB (Bradley & Sato, 2008). The decreasing signal-to-noise ratio in relation to the child’s age reveals that auditory recognition ability develops as the child matures. Speech recognition is roughly adult-like in 11–13-year-old children (Leibold & Buss, 2013). Despite the challenges posed by noise to listening conditions, many children spend their days in classrooms and facilities with noise levels of 60–78 dB (Rantala et al., 2011).

Because different kinds of noise interfere with word recognition in different ways, researchers have explored a range of different noise masks when testing the robustness of children’s word recognition. The spectrum of noise varies, for example, between babble noise and white noise, creating differing masking conditions to speech (Phatak & Allen, 2007; Phatak et al., 2008). Babble noise is noted to create so-called informational masking if it contains only a few individuals’ voices. In informational masking the listener’s perception and focus detecting the target words are challenged not only by the background noise but also by the competing words and content they hear in the masking noise (Freyman et al., 2004). However, the effect of informational masking on listening conditions fades away as the number of speakers exceeds ten; the competing information is harder to detect and it disturbs less (Freyman et al., 2004). Babble noise has been considered an effective masking noise for speech and non-speech stimuli in adults (Kozou et al., 2005). It has also been noted to impair immediate passage comprehension in normal-hearing children (Brännström et al., 2018). Babble noise is familiar to pre-school-aged children from daycare centers and playgrounds. Due to this familiarity, this kind of masking noise is suitable for different types of auditory recognition tasks with children.

Background noise may complicate the formation of consistent phonemic and lexical representations and blur phoneme recognition (Ahissar et al., 2006). Babble noise has been shown to affect different phonemes in various ways. In general, phonemes can be recognized more easily as they are presented with more energy above the level of masking noise (Meyer et al., 2013; Phatak & Allen, 2007). For example, sibilants appear to be highly noise-resistant consonants; background noise seems to have only little effect on their recognition (Moreno-Torres et al., 2017; Phatak & Allen, 2007). In one analysis, consonants could be divided into three groups according to their resistance to being obscured by babble noise (Moreno-Torres et al., 2017). The high noise-resistant group included phonemes with energy above the babble noise (e.g., /s/), the mid-resistant consonants contained energy in the first 4000 Hz (e.g., /r, l/), and the low-resistant consonants had limited energy across the spectrum (e.g., /t, θ, k, f, p/).

If a consonant is misheard, the listener confuses it most likely with another consonant that shares some features with the misheard consonant. Features that have been reported to lead to confusion between consonants are voicing and the place and manner of articulation (Benkí, 2003; Christiansen & Greenberg, 2012; Leibold & Buss, 2013; Meyer et al., 2013; Phatak & Allen, 2007). The consonants that are confused with each other have been noted also to share the same level of noise resistance (Benkí, 2003; Moreno-Torres et al., 2017). For example, two sounds that are both challenging to hear in noise are more likely to be confused than low and high noise-resistant consonants. Children have been noted to have mostly similar consonant confusion patterns in quiet and in speech-shaped noise as those in adults, but some differences are also reported. Young children (aged 4–5 years) faced more challenges in perceiving front placement (labial) stop consonants in a quiet environment than older children or adults (Nishi et al., 2010). Another kind of challenge in children’s consonant recognition was reported in a recent study by Brosseau-Lapré et al. (2020). They conducted a preliminary study of auditory recognition of an early-acquired phoneme (/p/) and a late-acquired phoneme (/ʃ/) in children (4;6–6;6) with and without SSD. They reported that all of the children recognized the early-acquired consonant more accurately than the late acquired one in a quiet environment. Although this result may be due to the acquisition age of the phoneme in question, also the fact that /p/ is a very frequent phoneme in English might have played a role. The evidence in what level the speech sound production and the auditory recognition impact each other is not clear (Hearnshaw et al., 2019). A good performance in auditory recognition does not necessarily lead to good production of a sound but may increase the likelihood to learn the pronunciation of the sound (Rvachew & Brosseau-Lapré, 2018). Thus, the maturity of speech sound perception may be connected to the order the speech sounds are acquired.

Investigating the auditory recognition ability of young children is a challenging task. Some guidelines have been presented for children aged 4–5 years (Vance et al., 2009). The task paradigm should be appropriate for age, containing age-appropriate vocabulary, and should not require any speech output. In the current study, even younger children (3;6 years) were evaluated. Thus, in addition to the previous recommendations, the task for younger children should be fast to complete and easy for children to learn. Simple equipment should be used (e.g., no additional loudspeakers or headphones). A closed-set auditory word recognition point-to-picture task of familiar words would fill these requirements. However, a possible ceiling effect should be taken into account when only familiar words are used. The presence of background noise challenges the listening. Thus, masking noise has been considered appropriate to use when the aim is avoidance of a ceiling effect (Vance et al., 2009). A type of test in which the stimulus gradually becomes more challenging (e.g., Hearing In Noise test; Nilsson et al., 1994) or a task where children judge the word to be similar or different from the previous word or if the word is pronounced correctly (e.g., Beving & Eblen, 1973; Hearnshaw et al., 2018; Vance et al., 2009) may be too challenging for children aged 3;6 years. To conclude, the assessment of word recognition ability of young children differs from that of older children and adults in many ways.

Specific characteristics of Finnish-speaking children’s phonology

In this study, the relation between auditory word recognition ability and development of phonological production is investigated in children at 3;6 years acquiring a Finno-Ugric language, Finnish. Finnish is a full-fledged quantity language (Suomi et al., 2008), i.e. the long (geminate) and short (singleton) phonemes in both consonants and vowels change the word meaning (e.g., [tuli, tu:li, tul:i] English: fire, wind, customs). Finnish is rich in vowels (8 monophthongs) and especially in diphthongs (18) but has a rather small consonant repertoire (13 indigenous consonants). In Finnish adults’ speech, the ranking of the indigenous consonants from most to least common is /s m t k n ʋ j p h l r d/ (Warren, 2001). The phoneme /ŋ/ is a Finnish sound but was not counted in Warren’s study.

Finnish children acquire phonemes in a mainly comparable manner to other languages. McLeod and Crowe (2018) reviewed paradigmatic development (phoneme inventories) in 27 languages, although not in Finnish. They reported that children at 3;6 pronounced correctly (reached a criterion) approximately 89% of the consonants in their native language, 97% of vowels, and 93% of all phonemes. In comparison, Finnish children mastered 85% of consonants, 100% of vowels and diphthongs, and overall 95% of phonemes at 3;6 (Aalto et al., 2020). The minor difference between the studies may be due to the small number of consonants but a large number of vowels and diphthongs in Finnish. Further, in Indo-European languages the trills have been reported to be late-developing phonemes, usually acquired before the fifth birthday (McLeod & Crowe, 2018). Also, in Finnish, the apical trill /r/ is often acquired the latest. Only 37% of children at 3;6 had acquired it in at least one position in a word (Aalto et al., 2020).

Regarding children’s ability to combine different phonemes based on the rules of their native language (i.e. phonotactic development), Finnish children exhibit phonological processes that are common in many languages. However, also language-specific processes are found. First, initial consonant omission is considered to be part of typical development in Finnish children (Saaristo-Helin, 2009; Savinainen-Makkonen & Salovaara, 2008), as it is for French (Wauquier & Yamaguchi, 2013), Hebrew (Keren-Portnoy & Segal, 2016), and Italian children (Vihman & Majorano, 2017). However, in some other languages, initial omissions were less likely to occur in typical development (Goldstein et al. (2004) [Spanish]; Petinou and Okalidou (2006) [Cypriot Greek]; Vihman and Croft, 2007 [English]). Second, a Finnish-specific process in children’s phonological development is the phenomenon that omissions in word-medial clusters disappear early, before the age of 3;0 (Saaristo-Helin, 2009). Instead of omitting a phoneme, Finnish children usually either assimilate the consonants in the cluster (e.g., [kas:i] for [kaksi] “two”) or lengthen the preceding vowel (e.g., [ka:hu] for [karhu] “a bear”) to avoid the pronunciation of a difficult cluster. In some other languages, word-medial cluster omissions are reported to be a typical pattern (e.g., Cohen & Anderson, 2011 [English]; Fox & Dodd, 1999 [German]; Gangji et al., 2014 [Swahili]; Grech, 2006; So & Dodd, 1995 [Cantonese]) existing in multisyllabic words in typically developing children up to seven years of age (James et al., 2008 [English]). One hypothesis explaining these language-specific phonological processes is the so-called geminate-pull effect (Vihman, 2010), which means that the word-medial geminate consonants pull the child’s focus to the middle of the word (Keren-Portnoy & Segal, 2016; Vihman & Majorano, 2017; Wauquier & Yamaguchi, 2013). The strength of the geminate-pull effect can be seen in Finnish, where young children are reported to favor the rhythm of the word (i.e. keeping the syllable structure intact) over the correct pronunciation (Torvelainen, 2007).

Regarding the connection between auditory word recognition and speech production of preschool children, the earlier research has mostly been analyzed in children with speech sound disorders (SSDs) (see recent review: Berti et al., 2020; Hearnshaw et al., 2019). Less is known about the relationship in typically developing children. This study aims to investigate the association between word recognition ability in background babble noise and phonological skills in a representative group of children aged 3;6 years, an age when phonological skills are still developing.

The present study

This study aimed to illuminate the relationship between auditory word recognition ability in babble noise and development of phonological production in children aged 3;6 years. In our earlier study (Aalto et al., 2020), we investigated associations between development of phonological production and previous and concurrent lexical ability in this same group of children. Now we explore whether phonological skills at age 3;6 are associated with auditory recognition ability at the same age. Studies that provide information on the association between auditory recognition ability in noise and emerging phonological skills in typically developing children are scarce. Furthermore, most studies concern English or other Indo-European languages. In this study, we investigate this association in a sample of Finnish children at 3;6. It is important to assess the development and associations between auditory recognition in noise and phonological ability since children tune to their native language already during their first year of life (Cheour et al., 1998; Werker & Tees, 1984), and mother tongue has been found to affect the auditory word recognition process (Dawson et al., 2016). The connections between auditory word recognition and phonological skills may reflect the development of the underlying phonological representations. Here, we investigate the association between word recognition ability in babble noise and development of phonological production first using the quantitative approach and then using descriptive analysis in order to get as comprehensive a picture as possible of this phenomenon. The detailed research questions are as follows:

Quantitative analysis:

1. Is there an association between auditory word recognition ability in babble noise and development of phonological production (paradigmatic development, i.e. phoneme inventories; phonotactic development, i.e. ability to combine different phonemes) in a representative group of children acquiring Finnish at 3;6?

2. How much do children’s auditory word recognition ability and phonological skills explain each other at 3;6 when the effect of background factors (gender, maternal education) are taken into account?

Descriptive analysis:

• (3) Do the following phoneme-related aspects affect word recognition in babble noise:

a. word includes a sibilant (/s/), b. word includes a late-acquired phoneme (/r/)?

• (4) Do mutual confusions occur between minimal pairs in which the distinctive phonemes differ by either place or manner of articulation?

The hypotheses are as follows: First, auditory word recognition ability in multi-talker babble noise and development of phonological production are associated with each other. This hypothesis is based on earlier findings of connections between auditory word recognition ability and development of phonological production in children four years and older with SSD (e.g., Brosseau-Lapré et al., 2020; Hearnshaw et al., 2019). Second, both auditory word recognition ability and the development of phonological production have some explaining value for each other when analyzed in regression models. This hypothesis is based on the assumption that phonological representations may be an underlying factor for both auditory recognition ability and phonological skills (e.g., Edwards et al., 2004; Metsala & Walley, 1998; Smith et al., 2006; Stoel-Gammon, 2011). Third, the words that include the sibilant /s/ are well recognized since sibilants have been noted to be resistant to babble noise in adult populations (e.g., Moreno-Torres et al., 2017; Phatak & Allen, 2007). Fourth, if a word includes the late-developing phoneme /r/, it creates more recognition confusion than other words. A recent study found that an early-developing phoneme was recognized better than a late-developing one by both typically developing children and children with SSD (Brosseau-Lapré et al., 2020). Fifth, most of the mutual confusion occurs between minimal pairs in which the distinctive phonemes differ by either place or manner of articulation. This hypothesis is based on previous studies showing that place and manner of articulation are especially vulnerable to confusion (Christiansen & Greenberg, 2012; Leibold & Buss, 2013; Meyer et al., 2013; Phatak & Allen, 2007).

This study is part of the norming research of the Finnish short-form version of the MacArthur Communicative Development Inventories (FinCDI-SF Study; project leader: the last author of the present article). The Ethics Committee of the University of Turku has approved the procedure of the FinCDI-SF study. All parents signed informed consent and received written feedback on their child’s language skills.

Methods

Participants

The subjects were 65 children (36 girls) whose language development was followed longitudinally from nine months of age onwards. This study focuses on the age point of 3;6. The public health nurses at Child Welfare Clinics in Turku invited the participating children and their parents to the study. The inclusion criteria were as follows: a full-term (born >37 gestational weeks) and healthy child raised in a Finnish-speaking monolingual family. Child-based exclusion criteria were a diagnosis (or suspicion) of cerebral palsy, intellectual impairment, autism spectrum disorder, or hearing impairment. In addition, family-based exclusion criteria comprised narcotic or alcohol addiction or severe mental health problems in one or both parents. The educational level of parents was as follows: 51% of mothers and 45% of fathers had been enrolled in university studies or had a master’s degree, 34% of mothers and 22% of fathers had been enrolled in undergraduate university studies or had a bachelor’s degree, 12% of mothers and 28% of fathers had been enrolled in vocational studies or had a vocational degree, and 3% of mothers and 1% of fathers had a high school diploma. Information on education level was missing for 3% of fathers.

Data collection

At 3;6 years of age, children’s auditory word recognition ability was tested using a closed-set picture-pointing auditory word recognition task (AWRT) in multi-talker babble noise. In the AWRT created for this study, children are presented with 36 familiar spoken words in multi-talker babble noise conditions in 5 sets of words. The target words were recorded in a quiet studio setting using Audacity software and a Dynamic MT-58 Vocal Microphone. The audio signal was stored on a computer hard disk. The speaker was a female Speech-Language Pathologist and researcher. She was instructed to speak using good voice quality, neutral tone of voice, medium speech rate, typical Finnish intonation (falling intonation), and at a 10 cm proximity to the microphone. The multi-speaker babble noise was layered at such a level that speakers’ language, phonemes, intonation, or content could not be identified and the possibility of informational masking was excluded (over 10 speakers). The noise contained both female and male speaking voices in continuous speech. The signal-to-noise ratio was set at 13 dB for each set of words separately using the set’s main energy level. Background babble noise was chosen to avoid a ceiling effect in the picture-pointing task (Vance et al., 2009). The chosen 13 dB signal-to-noise ratio has been reported to create good enough listening conditions for approximately grade 3 students but to generate challenges for younger children’s auditory word recognition (Bradley & Sato, 2008). Also, the pilot study indicated that this signal-to-noise ratio challenged word recognition but did not overly disturb the children’s ability to focus.

The task contained 36 colored, hand-drawn pictures of two-syllable words familiar to Finnish-speaking children. Target words were presented in five separate picture boards, two containing six words and three containing eight words. The reason for the uneven number of words on the boards was the rejection of the weakly recognized words during the pilot study described below. Boards 1–4 included a set of words that were phonetic neighbors to each other, i.e. differed by 1 or 2 phonemes from each other (e.g., [pala, raha, kana, saha, paha, pata, kasa, kala], piece, money, chicken, saw, bad, pot, pile, fish). The syllable structures were controlled, and only the initial and medial consonants changed. All of the words ended with an open syllable because it is a typical word structure in Finnish (Suomi et al., 2008). The 5th board contained four pairs of words that differed from each other by the length of a phoneme (i.e. geminate and singleton), thus changing the syllable structure (e.g., [mat:o–mato] “rug–worm”). Consonant frequency (Hakulinen et al., 2005) was taken into account when the auditory word recognition tasks were created. The word structures in the picture boards and examples are presented in Table 1. A complete list of words, including the rejected words, is provided in Appendix A.

Table 1. Description of the structure and descriptive statistics for the results of the auditory word recognition task (AWRT). N indicates the number of pictures in the picture board (PB). Initial C% and Medial C% indicate the percentage of correctly recognized target consonants. The percentage of correctly recognized initial/medial consonants were calculated from the total number of initial/medial consonants in PB in question. The percentage of correctly recognized words in each picture board (Total word recognition, TWR %) is also provided.

PB	N	Word structure;example	Mean (SD)	Min-max	Initial C%	Medial C%	TWR %
1	6	C/a/C:/i/; [tak:i] a coat	5.0 (1.1)	1–6	90	93	83
2	6	C/i:/C/i/; [hi:ri] a mouse	4.5 (1.3)	1–6	77	76	75
3	8	C/au/C/a/; [laula] sing	4.7 (1.5)	1–8	73	65	59
4	8	C/a/C/a/; [kala] a fish	7.1 (1.0)	4–8	91	92	89
5	8	geminate-singleton [mat:o–mato] a rug–a worm	6.6 (1.3)	3–8	65	88	83
Total	36		27.9 (3.2)	21–36	79	83	78

The familiarizing routine for the AWRT was as follows: Children were first familiarized with target words using elicited naming. If the child did not name the picture correctly, extra semantic cues were provided (e.g., for [ratti] “a steering wheel”: “this is in the car. The driver holds it and turns it from side to side”). Finally, the assessing SLP named the picture as a confirmation of the meaning. The speaker’s voice was then introduced. The babble noise was presented next and described to the child as a noisy daycare classroom or a party if a child did not attend daycare. Two example words were played as often as needed to ensure that the child comprehended the instructions, first in the quiet condition and then in babble noise. The sound pressure level was selected together with the child to match the child’s comfort level during the introduction. In the actual test phase, each word was presented once in babble noise and the child pointed to the picture that they associated with the presented word. The assessment was interrupted as many times as needed to ensure that the child was focused on the task.

A pilot study of AWRT was completed with 11 Finnish-speaking children attending daycare (7 girls) aged between 2;8 and 4;7 years. Informed consent was obtained from a parent of each participant. First, the familiarity of the words and the recognition of pictures were assessed. Only pictures recognized without a semantic cue by over 80% of the children were chosen for the final AWRT. The rejected words are listed in Appendix A. From the chosen images, all children named 28/36 pictures without any cues. All children recognized all selected pictures after a semantic explanation (e.g., a button: you have this on your coat). The trial masking levels were quiet condition and signal-to-noise ratios of 10, 13, and 15 dB. Statistical analyses were not completed due to the small number of subjects in the pilot sample and the varying age range of the children. When the 10 dB signal-to-noise ratio was tested, some children stated that they did not hear the words or that “the other people speaking” (babble noise) were too loud. All children recognized words from every picture board in a signal-to-noise ratio of 13 dB and were able to focus on listening. The pilot study participants did not participate in the current study.

Children’s phonological skills were assessed using the “Finnish Phonology Test” (FPT; Kunnari et al., 2012). In the FPT, a child names an object in a picture (90 pictures). The test analysis consists of a paradigmatic and a phonotactic part. The paradigmatic part assesses the word-initial, word-medial, and word-final consonant inventories of the 13 consonants and eight vowels in the Finnish language. For the word-initial consonant inventory, the maximum score is 12, for the word-medial consonant inventory 13, and for the word-final consonant inventory 2. The maximum score for the vowel inventory is 10 (8 vowels and 2 word-initial vowels). Each of the sounds in different positions that were tested, were tested with one lexical item. The total maximum of paradigmatic skills is thus 37.

The phonotactic part assesses the ability to combine phonemes in different syllables and word structures. The structures assessed are phoneme length (i.e. singleton/geminate phoneme), syllable structure (e.g., CVC, CVCC), word length (1–5 syllables), and ability to combine phonemes (diphthongs, consonant clusters, and phonemes in separate positions in a word). During the test performance, phonetic allophones of the phonemes /s/ and /r/ are ignored because there is only one sibilant and one trill in Finnish. The pronunciation of these sounds also varies among adult speakers. During the data collection, a trained SLP or SLP student scored the child’s performance based on the instructions provided in the FPT manual and based on the test protocol. The phoneme paradigm and phonotactic features were scored as either mastered (correct/similar to adult production) or not mastered (incorrect/different from adult pronunciation). With the non-mastered phonemes and structures, also the error type was noted (e.g., substitution, omission, compensatory lengthening, assimilation).

Data handling and analysis

In the analysis, the paradigmatic and phonotactic scores were summed to obtain a total score for phonological skills (max 164 points).

The associations between phonological skills (total score of FPT), paradigmatic skills, phonotactic skills, and AWRT total score were investigated using Pearson’s correlation coefficient values. The Mann-Whitney U-test was used to analyze gender differences in AWRT performance. To further investigate the association between word recognition ability in babble noise and development of phonological production, and to obtain information on the effect of background factors on this association, two multiple linear regression analyses were compiled. The outcome variable was either the total AWRT or FPT score. The predictor variables were total score of FPT or AWRT, maternal education level, and child’s gender. Statistical significance was assessed using ANOVA. The significance limit was set at p < .05.

In the descriptive analysis (N = 60; missing data from 5 participants due to missing detailed information in the AWRT recording sheet), the following features were investigated: word recognition based on phonemes in the initial and medial placement of a word, the five best- and least-recognized words, and the words that were misheard/confused the same way by at least 10 children.

Results

Data description

In AWRT, the mean value of correctly recognized words was 28 (SD 3.2, min-max 21–36). The variation of AWRT results is shown in Table 1. The word recognition rate varied in different picture boards from 59% (PB 2) to 89% (PB 4). At least one word was identified in every picture board by each child, and at least one child identified all of the words. Boys and girls did not differ from each other in AWRT score (Mann-Whitney U(29,36) = −0.17, p = .86). All target words with recognition rates are listed in Appendix A.

Table 2 presents the descriptive statistics for the FPT results and its parts (paradigmatic and phonotactic skills). Correct /r/ was pronounced by 20 of the 65 children in word-initial and word-medial placement. The most common substitution for /r/ was /l/ (27 children). Other substitutions were /w/ (2 children) and /ʋ/ (3 children). Information about the substituting consonant for /r/ was missing for 13 children.

Table 2. Descriptive statistics for the results of the finnish phonological test (FPT).

	Mean (SD)	Min-max
Paradigmatic score	34 (2)	25–37
Phonotactic score	108 (13)	52–127
FPT total score	142 (16)	77–164

Associations between auditory word recognition ability and phonological skills

A significant, moderate correlation between auditory word recognition and phonological skills was found (Figure 1). The correlation was comparable between auditory recognition ability and paradigmatic development (Figure 2) and between auditory recognition ability and phonotactic development (Figure 3). One child presented with a very weak score in phonology (<8th percentile). When the result of this child was excluded, the correlation between AWRT and FPT remained statistically significant (r(62) = 0.31, p = .014).

Figure 1. Scatterplot presentation for the association between the result of the auditory word recognition task (AWRT/total score) and the finnish phonology test (FPT/total score). The figure includes correlation coefficient value (r) with the degrees of freedom, and significance level (p).

Figure 2. Scatterplot presentation for the auditory word recognition task (AWRT/total score) and paradigmatic skills in the Finnish phonology test (FPT/paradigmatic score). The figure includes correlation coefficient value (r) with the degrees of freedom, and significance level (p).

Figure 3. Scatterplot presentation for the auditory word recognition task (AWRT/total score) and phonotactic scores in the Finnish phonology test (FPT/phonotactic score). The figure includes correlation coefficient value (r) with the degrees of freedom, and significance level (p).

Two multiple linear regression models were fitted to the data to predict FPT and AWRT scores and to take into account the possible effect of background factors. The first model was based on independent variables: maternal education level, child’s gender, and AWRT score, with the dependent variable being the FPT score. A significant regression equation was found (F(3.61) = 4.3, p = .008), with an adjusted R squared of 0.13. The children’s phonological skills were predicted as follows: the intercept was 87.8 and the AWRT slope was 1.62. The AWRT score was a significant predictor for the FPT score at 3;6 (ANOVA p < .01), while maternal education level (ANOVA p > .1) and child’s gender (ANOVA p > .1) were not significant. When the child with a very weak FTP result was removed, a significant regression equation remained (F(3.60) = 2.9, p = .04), with an adjusted R squared of 0.08. The children’s phonological skills were predicted as follows: the intercept was 103.1 and the AWRT slope was 1.12. The AWRT score was a significant predictor for the FPT score when the child with a very weak FPT score was excluded (ANOVA p < .05).

In the second linear regression model, the independent variables were maternal education level, child’s gender, and FPT scores, with the dependent variable being the AWRT score. A significant regression equation was found (F(3.61) = 3.85, p = .012), with an adjusted R squared of 0.12. The children’s AWRT results were predicted as follows: the intercept was 15.1 and the slope for FPT was 0.07 (p < .01). The FPT score was a significant predictor for the AWRT score at 3;6 years (ANOVA p < .01), while maternal education level and child’s gender were not significant.

Descriptive analysis of AWRT results

Comprehensive information on how children recognized words based on the initial and medial consonants is presented in Table 3. The words that included the phoneme /s/ were recognized overall the best (96%). Regarding the word-initial consonant, children recognized easiest the words that included the phoneme /s/ (99%) and the least the words with /r/ (71%). Regarding the word-medial consonant, children recognized the words with medial /k/ the best (99%) and medial /ʋ/ the least (50%).

Table 3. Consonants in the auditory word recognition task (AWRT) and their recognition rate in total and in initial and medial positions in the target words.

Target phoneme	Recognition of phoneme, %	Recognized in initial position, %	Recognized in medial position, %
/s/	96	99	92
/k/	93	87	99
/t/	89	88	89
/n/	88	94	82
/h/	83	85	80
/p/	77	83	70
/r/	73	71	75
/l/	72	73	70
/ʋ/	64	78	50

Recognition of the Finnish phonemes /m/ and /u/ is not included in this table because they created minimal pairs only in geminate-singleton word pairs (i.e. the phoneme quantity/length was the only changing feature while the phoneme stayed the same).

The five most- and least-recognized words in AWRT are listed in Table 4. Four of the five best-recognized words included the phoneme /s/. The least-recognized words included either a long vowel or a diphthong. Two of the five least-recognized words included the late-developing phoneme /r/.

Table 4. The five most- and the five least-recognized words and their recognition percentages. The translations of words in English are also presented. The analysis includes the answers from 60 children.

Best-recognized words	Recognition %	Least-recognized words	Recognition %
[saha] a saw	100	[pi:ri] a circle	50
[kis:a] a cat	98	[hi:vi] tiptoe	50
[si:li] a hedgehog	98	[kaula] a neck	40
[pata] a pot	98	[rauta] an iron	38
[kasa] a pile	95	[nauha] a ribbon	35

Word recognition confusions were analyzed. The words that were misheard in AWRT by ten or more children the same way, and their confusion pairs, are presented in Table 5. All the words were single-feature minimal pairs. The most common consonants in these pairs were /r-l/ (39x), /r-h/ (25x), /r- ʋ/ (20x), and /k-p/ (17x). The phonemes /r-l/ differ by the manner of articulation and the phonemes /k-p/ by place of articulation only. The phonemes /r-ʋ/ are both voiced continuants. The phonemes /r-h/ are continuants, but /r/ is voiced and /h/ is unvoiced.

Table 5. Words that were misheard in AWRT ten or more times the same way and their confusion pairs. The analysis includes the answers from 60 children.

Target word translation	Misheard as	Number of times misheard	Phoneme confusion
[nauha] a ribbon	[naura] laugh	25	/r/ for /h/
[hi:vi] tiptoe	[hi:ri] a mouse	20	/r/ for /ʋ/
[nap:i] a button	[nak:i] a hotdog	17	/k/ for /p/
[rauta] an iron	[lauta] a board	17	/l/ for /r/
[kisa] a race	[kis:a] a cat	13	/s:/ for /s/
[naura] laugh	[naula] a nail	12	/l/ for /r/
[pi:ri] a circle	[hi:ri] a mouse	11	/h/ for /p/
[u:ni] an oven	[uni] a dream	10	/u/ for /u:/
[naula] a nail	[naura] laugh	10	/r/ for /l/

Discussion

This study elucidates the associations between auditory word recognition ability in babble noise and phonological skills in monolingual Finnish-speaking children at 3;6 years of age. A moderate, significant positive correlation was found between phonological skills and auditory word recognition ability. The linear regression models showed significant but equally modest explaining value for both AWRT (12%) score and FPT (13%) score when the effects of gender and maternal education level were taken into account. The descriptive analysis of AWRT revealed that the sibilant /s/ was not concealed by the multi-talker babble, and children recognized well the words including /s/. The late-developing phoneme /r/ was involved in many word confusions. Consonants that lead to confusion in word recognition mostly shared either manner or place of articulation, although not always.

A significant and moderate association was found between auditory word recognition ability in the babble noise condition and phonological skills, even when the child with weak phonological skills was excluded, supporting the first hypothesis. These results of the representative group of Finnish children aged 3;6 years are comparable with previous studies completed with children aged 4;0 and older with SSD (e.g., Hearnshaw et al., 2019). Based on the current findings, auditory word recognition ability and phonological skills appear to be linked not only in children with SSD but also in typically developing younger children. The accuracy of phonological representations may mediate auditory recognition and phonological skills in both typically developing children and children with SSD – children with better phonological skills may also auditorily recognize words more precisely. This idea is in line with Smith et al. (2006), who hypothesized that more accurate phonological representations may lead to better pronunciation.

This study contributes information on the association between auditory and phonology skills in a Finno-Ugric language, Finnish, adding a new, little studied language to a growing body of evidence showing connections between auditory and phonological skills (e.g., Brosseau-Lapré et al., 2020; Hearnshaw et al., 2019). The association between auditory word recognition and development of phonological production seems to take place in different languages, although the ambient language has been noted to affect auditory discrimination (Dawson et al., 2016). Thus, the connection between auditory word recognition and phonological skills of children may be universal. However, more studies are needed in different languages and language families to verify whether this phenomenon occurs in multiple languages separately or can be generalized to all languages.

In the current study, a norm-referenced language-specific assessment method (FPT) was used to assess phonological skills. This test takes into account both phoneme inventories and children’s phonotactic skills. In a full-fledged quantity language, like Finnish, syllable structure plays such an important role that toddlers favor intact syllable structure over the correct pronunciation of phonemes (Torvelainen, 2007). Thus, phonotactic skills provide essential information on phonological skills in addition to the phoneme inventories. Based on the current results, both paradigmatic and phonotactic skills are equally accurate in predicting the auditory and phonological skills of Finnish children at 3;6. It would be interesting to examine whether auditory word recognition correlates more strongly with phonotactic skills than with paradigmatic skills in younger children since the geminate-pull effect has been noted to affect especially early phonological development (Keren-Portnoy & Segal, 2016; Vihman & Majorano, 2017; Wauquier & Yamaguchi, 2013).

The second hypothesis was that auditory word recognition ability and development of phonological production have some explaining value for each other when the effect of background factors is taken into account. This hypothesis was supported to some extent by the data showing modest, comparable explaining values in both directions. The accuracy of phonological representations may mediate both auditory word recognition skills and phonological development. Since connections between these skills are found not only in the SSD population aged 4;0 years and older (e.g., Hearnshaw et al., 2019) but also now in typically developing younger children (3;6 years), we assume that the mediating nature of phonological representations is seen earlier in typically developing children and later in children with SSD as a continuum. It is possible that when children grow older the connection between auditory recognition and phonological skills may not be found in typically developing children because the phonological skills are well-developed and the differences between individual children cannot be detected as easily as in younger children. However, the connection may be observed in children with SSD. Neither maternal education level nor child’s gender added predictive value to linear regression in this study. This may be due to the moderate sample size (N = 65) or the high education level in Finnish society. In previous studies, background factors (e.g., maternal schooling and gender) have correlated with a child’s language development (e.g., Eriksson et al., 2012; Kuvač-Kraljević et al., 2021).

The results support the third hypothesis; words that included the sibilant /s/ are well recognized. This finding is in line with earlier studies that have noted high-energy fricatives as being the most noise-resistant phonemes in the adult population (Meyer et al., 2013; Moreno-Torres et al., 2017; Phatak & Allen, 2007). In the present study, four of the five best-recognized words included /s/, even though the words themselves may not have been the most frequent in the child’s vocabulary ([saha] a saw; [si:li] a hedgehog; [kasa] a pile). And vice versa, none of the least-recognized words, or word-pairs most confused with each other, included the phoneme /s/. Thus, our results suggest that the high-frequency sibilant /s/ provides information for auditory discrimination also in those cases in which the meaning of the word may be less familiar to a child.

The fourth hypothesis about the late-acquired phoneme /r/ being included in many of the recognition confusions was supported by the descriptive analysis. Our result is in line with an earlier finding that children recognize an early-acquired phoneme easier than a late-acquired one (Brosseau-Lapré et al., 2020). The phoneme /r/, an apical trill, was mastered only by 20 of the 65 children in this study. It is possible that most of the children had a less accurate phonological representation of this phoneme since they did not yet have a motor scheme for the phoneme. The less accurate representation may, in turn, have challenged the auditory recognition as well. The finding that words including /r/ were less well recognized in noise than words including other consonants more concealed by noise than /r/ (Moreno-Torres et al., 2017) suggests that late-acquired sounds do indeed challenge word recognition in young children.

The fifth hypothesis that consonant confusion occurs between words where the distinctive phonemes differ by either place or manner of articulation thus being single-feature minimal pair words was supported by the data. The most common consonant pair leading to word recognition confusion was /r-l/. This result is in line with previous studies (Benkí, 2003; Meyer et al., 2013). Furthermore, both /r/ and /l/ have been described to have the same kind of noise resistance level (Moreno-Torres et al., 2017). In our study, the phoneme /l/ was noted to be the most common substitution for the phoneme /r/. This may be due to the same place of articulation of these two phonemes. Also, the phonemes may share some acoustic similarities that make /l/ a plausible substitution for /r/.

The possible effect of the place of articulation as a confusion source was noted in stop consonants /k-p/, as reported also in previous studies (Christiansen & Greenberg, 2012; Leibold & Buss, 2013; Phatak & Allen, 2007). However, not all consonant pairs leading to word recognition confusion in this study had phonemes that share either the place or the manner of articulation. The pair /h- r/, which was a common pair for confusion in this study, does not share articulation features, except for both being continuants. However, both of these phonemes may be challenging to hear for different reasons. The phoneme /h/ contains a low level of energy that may challenge its recognition (Moreno-Torres et al., 2017). The /r/ is a late-acquired phoneme, which may hinder its recognition (Brosseau-Lapré et al., 2020). The set of pairs that tended to be more confusable can be explained at least partially by the nature of the closed-set task where all phonemes are not presented to a similar extent. Same–different or correct–wrong recognition tasks would have allowed more control over the stimuli (e.g., Beving & Eblen, 1973; Hearnshaw et al., 2018), but these kinds of tasks were found to be too difficult for this age group.

Strengths and limitations

A strength of this study is the relatively large and representative group. Another strength is that it describes the relationship between phonological skills and auditory word recognition ability at an age when children are still typically acquiring their phonological skills. In addition, information on an understudied language, Finnish, is provided. Examining languages besides English is essential to gain a full picture of universal and language-specific features in speech and language development. Lastly, a clear strength is the use of a language-specific normed test to assess phonological development. However, the lack of a standardized auditory word recognition test is a limitation. Since a standardized auditory word recognition test was missing, a new task was created. Furthermore, the closed-set picture recognition used in the AWRT may have challenged children’s lexical skills. However, this effect was diminished by looking at the pictures together with the child and naming them before the task began. Because the task did not aim to evaluate children’s lexical skills, only well-recognized pictures were used and a semantic explanation was given to support word recognition. In addition, the AWRT did not contain all of the Finnish phonemes equally in all contexts due to the nature of the closed-set picture-pointing task. The task thus provided only partial information on consonant confusion patterns and noise resistance features of the phonemes.

Clinical implications

Background noise can mask fine phonetic details, thereby disturbing the formation of phonological representations also in typically developing children. The background noise levels in children’s daily environments should therefore be controlled to ensure good listening conditions and to enable the acquisition of fine-tuned phonological representations. Our findings showed that children’s phonological skills and auditory word recognition in babble noise are associated with each other. Thus, children with strong phonological skills may have better auditory recognition ability also in day-to-day noisy conditions than children with weak phonological skills. This should be taken into consideration in children’s daily environments such as daycare centers.

Conclusions

This study showed a moderate and significant association between auditory word recognition ability and phonological skills in typically developing children at 3;6 years of age. These results from Finnish, a Finno-Ugric language are in line with previous findings from Indo-European languages. Noisy environments in children’s everyday lives may influence children’s word recognition differently based on their current phonological skills.

Acknowledgments

We thank visual artist S. Lehti for drawing the pictures for the AWRT and allowing us to use them in this study. We also thank all families and other study participants.

Disclosure statement

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

Appendix 1.

Words used in the Auditory Word Recognition Task, their recognition percentages, and translations to English. The rejected words (i.e. words not included in the AWRT) are also presented.

PB 1	%	PB 2	%	PB 3	%
[rat:i] a steeringwheel	93	[si:li] a hedgehog	98	[kauha] a scoop	88
[nak:i] a hotdog	92	[hi:ri] a mouse	87	[lauta] a board	72
[tak:i] a jacket	88	[vi:si] five	82	[naula] a nail	70
[lak:i] a cap	87	[vi:li] a yogurt	73	[laula] sing	62
[kat:i] a kitty	82	[hi:vi] tiptoe	50	[naura] laugh	58
[nap:i] a button	68	[pi:ri] a circle	50	[kaula] a neck	40
				[rauta] an iron	38
				[nauha] a ribbon	35
rejected words in PB 1		rejected words in PB 2		no rejected words
[tat:i] ‘a mushroom’		[hi:li] a coal
[tap:i] ‘a peg’		[vi:ri] a pennant
[rak:i] ‘a dog (mutt)’		[ti:li] a brick
PB 4	%	PB 5	%
[saha] a saw	100	[kis:a] a cat	98
[pata] a pot	98	[pal:o] a ball	93
[kasa] a pile	95	[mat:o] a rug	93
[kala] a fish	95	[mato] a worm	90
[kana] a chicken	92	[kisa] a race	77
[raha] a money	80	[palo] a fire	77
[pala] a piece	68	[uni] a dream	72
[paha] bad	77	[u:ni] an oven	60
no rejected words		rejected words in PB 5
		[kuka] who (pairing geminate: [kuk:a] flower)