Importance of speech intelligibility and factors influencing

Considering that intelligible speech is required for daily interactions, studies have shown that intelligibility is linked with social well-being. For example, intelligibility and hearing peers’ attitude towards HI children were shown to correlate: as the intelligibility decreased, peers rated the personal qualities (cognitive and emotional-behavioural factors) of the speaker more negatively (Most et al., 1999). Low intelligibility also affected the speaker’s psychosocial competences. Parental reports revealed that several measures such as anxiety and social withdrawal correlated with intelligibility (Freeman et al., 2017). Thus, low intelligibility can lead to social isolation and lower psychosocial outcomes.

Considering the social consequences, reaching intelligible speech is an important milestone in the development of children. For NH children, the benchmark of reaching intelligible speech is set at around four years of age (Chin & Tsai, 2001; Weiss, 1982). Children who have reached this age and whose transcribed connected speech is less than 66% intelligible are candidates for speech and language therapy (Gordon-Brannan & Hodson, 2000).

Although there is consensus on the importance of intelligible speech, the aspects that contribute to the intelligibility of an utterance are less clear. In the following paragraphs, four factors will be discussed separately. Obviously, intelligibility depends on the speaker (§1.4.2.1) and the listener (§1.4.2.2). In an experimental or diagnostic setting, two additional factors contribute: the type of utterance (§1.4.2.3) and the type of measurement (§1.4.2.4).

1.4.2.1 Role of the speaker

When considering intelligibility merely as a product of the speaker, intelligible speech requires a combination of “prerequisite speech perception abilities to learn and understand speech, the linguistic knowledge to plan and execute spoken utterances, and the motor control abilities to articulate meaningful sentences” (Montag et al., 2014: 2332). In other words, intelligible speech requires – among others – accurate speech and language production.

With respect to speech production, intelligibility is oftentimes considered as “the product of a series of interactive processes as phonation, articulation, resonance and prosody” (De Bodt et al., 2002: 284). Of these aspects, articulation is the most obvious and allegedly the most important contributor to intelligible speech (Baudonck et al., 2010b). It is however a misconception that deviances in articulation automatically lower intelligibility. In a study of Ertmer (2010) on the relation between intelligibility and word articulation, it was found that the articulation quality of words cannot reliably predict a child’s connected speech intelligibility. This can be due to the fact that measurements of particular

characteristics of speech such as articulation can objectively be measured, yet these measures do not necessarily correspond to how listeners perceive children’s overall speech intelligibility (Uchanski & Geers, 2003).

Moreover, intelligible speech requires more than just adequate speech. As Ferguson (1981: 1) stated: “child’s intelligibility depends upon his ability to transmit the language code; a task which includes knowledge and proficiency with language – the expression of ideas through words”. In other words, linguistic aspects possibly affect speech intelligibility. For example, phonological distance measures, i.e. measures that compare the child’s production with the target at the phoneme level, have been shown to be correlated with speech intelligibility (Sanders & Chin, 2009).

Morphological and syntactic features such as the length of the utterance, the position of words and the syllabic structure have also been suggested to affect perceived intelligibility (Chin & Tsai, 2001; Gordon-Brannan &

Hodson, 2000; Weston & Shriberg, 1992). Again, there does not seem to be a direct relation between intelligibility and a specific linguistic aspect.

Rather, all aspects affect intelligibility interactively (Van Lierde et al., 2005). Also, speech and language aspects are suggested to interact. More specifically, listeners seem to use compensating mechanisms. For example, if the speech of an individual is particularly affected, listeners will rely more on linguistic components in order to decode the utterance of the speaker.

Vice versa, listeners will rely more on clear speech if the speaker demonstrates linguistic errors (Osberger, 1992).

1.4.2.2 Role of the listener

As discussed in the paragraph on the speech chain, the question whether an utterance is considered (un-)intelligible, is determined by the speaker as well as the listener (Kent et al., 1994; Rietveld & van Heuven, 2016). In other words, when listeners are required to judge children’s (disordered) speech, intelligibility is “a measure of how well a listener can

‘tune in to’ a deviant speech pattern” (Ferguson, 1981: 11). Individual characteristics of listeners thus have an influence on the perceived intelligibility. For example, in order to have a successful conversation, listeners have to know the same language as the speaker. Another example:

a hearing impairment may hamper a swift conversation (Flipsen, 2008).

Another influencing factor is the amount of listener experience. More specifically, studies often suggest that an increased amount of experience with a specific type of speech facilitates its intelligibility (Klimacka et al., 2001; McGarr, 1983; Monsen, 1978; Munson et al., 2012; Osberger, 1992).

For example, listener groups such as primary school teachers or speech and language pathologists have had more experience with (HI) children’s speech and are thus expected to rate the speech intelligibility of these children to be higher than inexperienced listeners.

There are two types of studies in which the variability in listeners is taken into account. On the one hand, there are studies in which the listeners are already experienced with a particular type of speech prior to the experiment (amongst others: Flipsen, 1995; McGarr, 1983; Munson et al., 2012). On the other hand, there are studies in which listeners are familiarised with a particular type of speech as part of the study (Beukelman & Yorkston, 1980; Borrie & Schafer, 2015; Ellis & Beltyukova,

2008; Ferguson, 1981; Tjaden & Liss, 1995). In this dissertation, the judgements of listeners who have experience with the speech of (HI) children will be compared to those of inexperienced listeners.

Studies comparing experienced and inexperienced listeners do not show univocal results. On the one hand, some studies found that the intelligibility judgements of experienced listeners were consistently higher than those of inexperienced listeners (McGarr, 1983; Monsen, 1978), that experienced listeners used context more efficiently (Osberger, 1992) or made more reliable and valid judgements (Klimacka et al., 2001; Munson et al., 2012). On the other hand, some studies did not find this difference between inexperienced and experienced listeners. They reported that both listener groups performed similarly (Flipsen, 1995; Gillis, 2013; Grandon, 2016; Mencke et al., 1983). Thus, in these studies, the experience that for example speech and language pathologists have gathered in their professional career did not lead to a higher proficiency in judging HI children’s speech.

There are several possible explanations for these contradicting results. One possibility is related to the speech intelligibility of the children. Some studies suggest that the children’s intelligibility might be so low that – even with considerable experience – decoding the speech is sheer impossible (Mencke et al., 1983; Svirsky et al., 2007). The opposite scenario is also possible: the intelligibility is so high that experienced as well as inexperienced listeners perceive perfectly what the speaker intended to say.

Moreover, there are several reasons as to why experienced listeners rate an utterance to be highly intelligible. First, it can be expected that experienced listeners are indeed better at decoding children’s speech and thus judge their speech intelligibility more correctly or more consistently.

However, it can also be expected that experienced listeners overestimate children’s speech because of their experience with a wide variety of speech.

Since they hear different degrees of disordered speech on a daily basis, their perception of unintelligible speech might have shifted: whereas inexperienced listeners may find one type of speech fairly unintelligible, the experienced listeners may appreciate it better because they – consciously or unconsciously - compare it to an even less intelligible type of speech. Thus, a high intelligibility score is either due to the result of better speech decoding skills or to the overestimation of the experienced listener (Beukelman & Yorkston, 1980; Munson et al., 2012).

1.4.2.3 Type of utterance

Imitated/read speech vs. spontaneous speech

Concerning the speech samples that are used for intelligibility assessments, previous studies have explored several options. First of all, one of the main distinctions that has to be made is whether the speech samples originate from imitated or read speech or from spontaneous speech. In imitated speech tasks, speakers are either prompted to imitate speech utterances that are demonstrated by the examiner or to read aloud utterances. Prior to the assessment, the researcher selects the speech samples that the speaker has to imitate or read. Next, listeners judge the speech samples by means of, for example, orthographic transcriptions. The

fact that the researcher has written transcripts, i.e. a model, of the utterances that were imitated by the speakers, has the important advantage that the number of correctly identified words in the transcriptions of the listeners can easily be calculated. Alternatively, the speech intelligibility assessment is performed by means of spontaneous speech samples.

Spontaneous speech can be elicited by pictures, in informal conversations or by story (re)telling. From these speech recordings, speech samples are extracted and presented to the listeners. The main advantage is that spontaneous speech is more representative for day-to-day speech and thus has a higher ecological validity. The main disadvantage is that, in contrast to imitated speech, there is no model of the speech utterances. In other words: spontaneous speech utterances are unpredictable and uncontrollable to a large extent. The lack of a model also has implications on the scoring procedure, since the utterances of the speaker cannot be compared with a prewritten transcript.

Words vs. longer speech samples

The second factor concerning the type of utterance is related to the linguistic unit. The spectrum of linguistic units goes from small units such as words up to longer stretches of speech. All types of utterances imply different elicitation tasks for the speaker. For example, at the word level, the most common tasks are a picture naming task or reading a list of words (Huttunen, 2008; Löhle et al., 1999; Vieu et al., 1998). At the sentence level, there are several preset imitation tasks that reoccur in several studies. The most frequently used sentence tasks are the Beginner’s Intelligibility Test

(Monsen, 1978) and the McGarr Sentence Intelligibility Test (Mcgarr, 1981).

A combination of both BIT and Monsen Sentence Test is often performed when the intelligibility of illiterate children (i.e. BIT) as well as older children (i.e. Monsen Sentence Test) is assessed in one study (Ertmer, 2010; Miyamoto et al., 1996; Svirsky et al., 1999; Svirsky et al., 2000a). In these tasks, the examiner reads the sentence aloud and the speaker is instructed to repeat the sentence or, if the children are old enough, they will read the sentences themselves. In some cases, the sentences are illustrated by pictures or objects (Osberger et al., 1994). For longer stretches of speech, the speech samples are extracted from speech recordings of informal conversations. Thus, whereas the first two linguistic units, i.e. words and sentences, mostly originate from an artificial experimental setting, the longer sequences of speech are mostly extracted from spontaneous conversations. In the present dissertation, short sentences as well as longer stretches of speech will be investigated by means of speech samples extracted from spontaneous speech.

1.4.2.4 Type of measurement

In the previous paragraphs, it was already established that studies on intelligibility vary on numerous factors. As a result, there is a large range of different types of measurement. Depending on the speaker, the listener and the type of utterance, (1) the task of the listener and (2) the quantification of the intelligibility score can be assessed differently.

Concerning the listener’s task, the most suited approach mostly depends on the type of utterance. For imitated or read speech, transcriptions as well as rating scales are commonly used. More

specifically, at the word or sentence level, listeners are mostly instructed to transcribe (orthographically or phonetically depending on their expertise) the utterance of the child (Chin et al., 2012; Chin et al., 2003; Ertmer, 2007;

Mencke et al., 1983; Monsen, 1978; Montag et al., 2014; Osberger et al., 1994; Svirsky et al., 2000a; Tobey et al., 2004). Longer stretches of imitated speech, for example a read text, can in principle also be transcribed, but this is an extremely time-consuming task for the listeners. Therefore, in this context, the listeners are most likely asked to judge the read text on rating scales. However, it should be noted that read texts are more frequent in the assessment of adults’ speech rather than children’s speech.

Therefore, they are not further discussed. Spontaneous speech is almost exclusively judged with rating scales (AlSanosi & Hassan, 2014; Bakhshaee et al., 2007; De Raeve, 2010; Ellis & Pakulski, 2003; Toe & Paatsch, 2013;

Van Lierde et al., 2005), because (1) they mostly investigate long stretches of spontaneous speech and it would be unfeasible to transcribe them and (2) because transcribing spontaneous speech has some important limitations that also affect the calculation of the intelligibility score.

Generally, the researcher’s approach to calculate the intelligibility score is fairly straightforward. For transcriptions (of imitated or read aloud speech), the intelligibility score is mostly the percentage correct averaged by the number of listeners. For rating scales, the ratings of the listeners are averaged. For the transcription of spontaneous speech, however, the calculation of the percentage correct is problematic because there is no model of the utterance. Hence, the researcher cannot compare the transcription of the listener to the “original” utterance that the speaker had

transcription of spontaneous speech. In the studies of Flipsen and Colvard (2006) and Lagerberg et al. (2014), listeners were instructed to transcribe the spontaneous speech samples orthographically and to indicate unintelligible syllables with for example “0”. For the calculation of the intelligibility score, Flipsen and Colvard (2006) tested different approaches to estimate the number of unintelligible words based on the number of unintelligible syllables. For example, one measure was based on the assumption that (English) child speech contains 1.25 syllables per word.

The number of unintelligible words in this case is thus the number of unintelligible syllables divided by 1.25. In the study of Lagerberg et al.

(2014), the conversion of syllables into words was not performed. Here, the intelligibility score was determined by dividing the number of intelligible syllables by the total number of syllables. However, both studies approach intelligibility in a very coarse manner and heavily rely on the judgements of each listener. Thus, there is a need for an intelligibility measure that is practicable and more explicit than estimations of the number of (un-)intelligible syllables. In this dissertation, an alternative approach to assess the intelligibility of spontaneous speech is explored (chapter 2).

1.4.3 Studies on the intelligibility of hearing-impaired children