Introduction

2.1.1 Intelligibility in hearing-impaired children

Reaching intelligible speech is an important milestone in children’s speech and language development. In the present study, intelligibility is conceptualised as a process in oral communication in which the listener identifies linguistic units in the speech stream of the speaker (Freeman et al., 2017; van Heuven, 2008; Whitehill & Ciocca, 2000). For typically developing children, speech that is intelligible for all listeners is estimated to occur around the age of four (Chin & Tsai, 2001; Chin et al., 2003; Weiss, 1982). Reaching this milestone in due time is also important from a diagnostic point of view: when one third of the continuous speech of a four-year-old cannot be transcribed correctly by others, this child is a candidate for speech therapy (Gordon-Brannan & Hodson, 2000). Intelligibility is also linked with psychosocial competences: children with low intelligibility have fewer social interactions with (hearing) peers, which may result in social isolation (Freeman et al., 2017; Most et al., 2012).

In the current study, the speech intelligibility of children with a cochlear implant (CI) is investigated. This hearing device can partially restore a severe-to-profound sensorineural hearing loss, but the period of auditory deprivation before implantation, the remaining hearing loss after implantation, and the degraded signal coming from the device’s sound processor have implications for these children’s speech and language development (O'Donoghue, 2013). When assessing the overall outcome of the speech production of children with CI, speech intelligibility measures are considered “the gold standard for assessing the benefit of cochlear implantation” (Chin et al., 2012: 356).

Studies have shown that following implantation the intelligibility of children with CI is still significantly lower than that of NH peers but it steadily increases thereafter. For example, in the study of Chin and Tsai (2001), the speech of four-year-old children with CI – implanted around the age of three – is on average 12.7% intelligible for inexperienced listeners. At this age, the speech intelligibility of NH children reaches (near) ceiling levels (Chin & Tsai, 2001; Weiss, 1982). Although speech development is initially delayed, an increasing amount of studies shows that some children with CI catch up with NH peers (Boons et al., 2013;

Geers & Nicholas, 2013; Habib et al., 2010; Nicholas & Geers, 2007; Wie, 2010). Here, two factors are of particular importance: age of implantation and length of device use (Flipsen & Colvard, 2006). Concerning the age of implantation, Habib et al. (2010) found that the speech intelligibility of primary school aged children implanted prior to 24 months is 15-18% more intelligible than that of primary school aged peers implanted between 24 and 36 months of age, when tested at a comparable chronological age.

Similarly, children implanted around the age of five reach only 68%

intelligibility after seven years of device use (Peng et al., 2004), whereas children implanted around the age of two with the same amount of device experience reach near ceiling scores (Montag et al., 2014). Thus, early implantation and prolonged device experience appear to lead to higher levels of speech intelligibility in children with CI (Chin et al., 2012; Chin &

Kuhns, 2014).

2.1.2 Intelligibility testing: imitated vs. spontaneous speech tasks Studies on children’s speech intelligibility can be divided into two categories according to the type of speech material that is collected: studies using read or imitated speech and studies in which the children’s speech is less pre-set or planned ahead. In this study, all speech that does not originate from read or imitated speech tasks will henceforth be referred to as spontaneous speech. Imitated or read speech tasks differ from spontaneous speech tasks in how the speech samples are elicited from the children, how the listeners are instructed to judge the speech samples and how the intelligibility score is calculated.

The vast majority of studies uses read or imitated speech (i.a.

Bharadwaj & Assmann, 2013; Castellanos et al., 2014; Chin et al., 2012; Chin

& Kuhns, 2014; Freeman et al., 2017; Khwaileh & Flipsen, 2010; Montag et al., 2014). Children are simply prompted to repeat utterances that were read by a researcher. Using imitated or read aloud speech has several advantages over spontaneously produced speech. For instance, there is a large amount of control over the stimuli. This controlled approach can be

useful for speech and language pathologists who use the results of the intelligibility test as a starting point for their child-specific speech therapy (Flipsen, 2006). Moreover, using the same imitated or read speech task leads to comparable speech samples between participating children.

On the other hand, using an imitated or read speech task has several disadvantages as well. Especially with hearing-impaired participants it is not always clear if errors in the imitation of stimuli are the result of not correctly perceiving the utterance acoustically (Khwaileh & Flipsen, 2010).

Indeed, errors may have nothing to do with the child’s linguistic or speech skills since the child only repeated what he/she (erroneously) thought that the utterance was. Furthermore, even young children act differently in a therapeutic setting than in spontaneous interactions. Because they are aware of the test environment and the fact that the researcher is observing their speech, children may adjust their speech by, for example, speaking more clearly (Ertmer, 2010). Because of these disadvantages, examining intelligibility through read or imitated speech at for example word level has been suggested to be “rather poor predictors of scores for connected speech and everyday performance with hearing aids” (Cox & McDaniel, 1989: 347).

Spontaneous speech is an alternative for read or imitated speech in assessing speech intelligibility. The most important advantage of spontaneous speech is its greater ecological validity. In other words, spontaneous speech is more comparable to everyday informal speech.

Despite this major advantage, only few studies use spontaneous speech for assessing children’s speech intelligibility (i.a. De Raeve, 2010; Lejeune &

Demanez, 2006; Tye-Murray et al., 1995; Van Lierde et al., 2005). This may be due to the lack of control over the speech sample: whereas in read or

imitated speech, the researcher decides on the words or utterances that the child is asked to read or imitate, this control by the researcher is far less outspoken in spontaneous speech. The child decides what to say. As a consequence, samples are not as comparable between speakers.

2.1.3 Measuring speech intelligibility scores

With respect to intelligibility measurements, the most commonly used tasks are rating scale tasks and transcription tasks. There are different types of rating scales, such as numeric rating scales, i.e. scales without a verbal description of predefined categories (i.a. AlSanosi & Hassan, 2014;

Habib et al., 2010; Ozbic & Kogovsek, 2010; Tseng et al., 2011), but probably the most frequently used rating scale is the Speech Intelligibility Rating (SIR) developed by Cox and McDaniel (1989) (i.a. Calmels et al., 2004; De Raeve, 2010; Flipsen, 2008; Lejeune & Demanez, 2006; Toe & Paatsch, 2013). The SIR requires that participants score each stimulus on a five-point scale with a verbal description for each score, ranging from unintelligible speech even for an adult familiar with the child to completely intelligible for all listeners. Rating scales such as the SIR offer a valid indication of the children’s speech intelligibility (AlSanosi & Hassan, 2014;

Fang et al., 2014; Flipsen, 2008), especially for assessing the intelligibility of very young (preschool) children with CI or children implanted at a later age (e.g., late kindergarten). However, children implanted at an early age (< 24 months) soon reach ceiling scores on the SIR, especially when their spontaneous speech was assessed at primary school age, i.e. roughly between the ages of 6 and 12 (Baudonck et al., 2010b; De Raeve, 2010; Toe

& Paatsch, 2013). This score indicates that – according to the SIR – their speech is intelligible to all listeners. Thus, it seems that after some years of device use, the speech of early implanted CI children has become “too intelligible” for the SIR to further differentiate. Generally, rating scales (independent of their type) are too coarse to measure more fine-grained differences between the intelligibility of CI children’s speech and the speech of NH children. In other words: children may be considered to be very intelligible according to rating scales but there may still be unintelligible parts in their speech (Miller, 2013).

Compared to rating scale tasks, transcriptions take a more fine-grained approach towards measuring speech intelligibility. In this type of task, listeners are asked to phonetically or orthographically transcribe the utterances. With respect to the listener’s task, transcription tasks can be identical for imitated or read speech and spontaneous speech. However, the way in which the intelligibility score is calculated, differs considerably for the different types of speech. With imitated or read speech, a simple comparison of the target words and the transcribed words is made, resulting in a percentage correct score. With spontaneous speech, calculating the intelligibility score of a transcription is not straightforward.

As was already mentioned in the previous section, the child decides what to say in a spontaneous speech task. This has some important theoretical and practical implications for how the intelligibility is calculated. Because only the speaker knows the exact intended utterance, there is no certainty about the correct transcription of the utterance without having the speaker confirm the transcriptions on hindsight (Flipsen, 2006; Flipsen & Colvard, 2006; Lagerberg et al., 2014). In other words, spontaneous speech

utterances do not have a model to compare the transcription with. As a result, the percentage correct cannot be calculated.

A possible alternative is to instruct the listeners to indicate the number of unintelligible syllables in their transcriptions (Flipsen, 2006).

However, this method comes with several limitations. Firstly, it is a difficult and time-consuming task for the listeners to determine the number of (un-)intelligible syllables, especially in long stretches of unintelligible speech. Secondly, there are several approaches to calculate the number of unintelligible words in the number of unintelligible syllables (Flipsen, 2006). Another consequence of the lacking model is that a post hoc qualitative analysis on the differences between the model and the transcription is not possible. Moreover, comparable to other studies on intelligibility, the scores are based on the judgements of only a few judges.

Inter- or intrajudge variability can thus affect the final intelligibility score to a large extent.

2.1.4 Aims of this study

Since transcriptions of spontaneous speech are difficult to judge in terms of correct or incorrect, the present study judges the transcriptions in terms of the degree of (dis-)agreement between listeners’ transcriptions using entropy scores. Entropy was originally developed in information theory (Shannon, 1948) as a measure that expresses the degree of disorder (“chaos”) in data. In linguistic research, entropy measurements were already employed for investigating the mutual intelligibility of two closely related languages such as Swedish and Danish (Frinsel et al., 2015; Moberg

et al., 2007). When Swedish and Danish cognates are aligned at a phoneme (or grapheme) level, identical phonemes in both languages imply agreement, and hence “low disorder or chaos”. Different phonemes in both languages imply disagreement, and hence, more “disorder or chaos”. In the former case, entropy is low, and in the latter entropy is higher. Judged over all the phonemes of two cognates, the final (average) entropy score predicts the mutual intelligibility of the two cognates.

In the present study, the intelligibility of primary school aged NH and CI children’s spontaneous speech is investigated through transcriptions and entropy calculations. Until now, spontaneous speech intelligibility was mostly investigated by means of numerical scales which are quite coarse so that early implanted children with several years of device use already reach (near) ceiling scores (Baudonck et al., 2010b; Chin et al., 2012; Chin & Kuhns, 2014). Since the use of transcriptions as well as spontaneous speech have both been suggested to lead to a lower intelligibility score (Khwaileh & Flipsen, 2010; Osberger, 1992), we expect to find (perhaps small) deviances between NH and CI children. Moreover, similar to other CI literature, we hypothesize that the entropy scores will show greater variability between subjects with CI (Castellanos et al., 2014;

Freeman et al., 2017; Montag et al., 2014; Nittrouer et al., 2014; Peng et al., 2004; Yanbay et al., 2014; Young & Killen, 2002). Therefore, the analysis will proceed in two steps. First of all, the intelligibility of children with CI and NH will be compared at a group level. Secondly, the individual variation between the children will be investigated.