Exploring the Relationship Between Mental Boundary Strength and Phonetic Accommodation

Abstract

We motivate a possible relationship between the psychological concept of mental boundaries and the linguistic phenomenon of phonetic accommodation, proposing that thinner boundaries may indicate a greater disposition to phonetically adapt to an interlocutor. To enable research on this relationship with German speakers, we translated the English short version of the Boundary Questionnaire (BQ-Sh), an established instrument for measuring the strength of mental boundaries, and demonstrated that the resulting German adaptation (BQ-Sh-G) can be used equivalently to the BQ-Sh. As the Big Five personality traits have previously been considered in research on both mental boundaries and phonetic accommodation, we explored the relationship between the BQ-Sh-G and the NEO-Five Factor Inventory. Consistent with previous literature, the BQ-Sh-G score correlated positively with Neuroticism and Openness, as well as negatively with Conscientiousness. We collected BQ-Sh-G scores from participants of an experiment on phonetic accommodation in a human–computer interaction context, specifically investigating the realization of the word ending 〈-ig〉 and the intonation of wh-questions in German. The analysis revealed a tendency for thicker mental boundaries to correspond with more convergence to 〈-ig〉 variants. Taking into account the results of previous work exploring the influence of the Big Five on the same data, we conclude that speakers may accommodate to different types of phonetic features depending on different personality traits. We encourage future work to investigate this further, while also exploring the predictive potential of the boundary construct with respect to a general disposition to phonetic accommodation, that is, examining a large number of phonetic features simultaneously.

1 Introduction

Phonetic accommodation refers to articulatory and vocal adjustments made by a speaker in response to being exposed to speech input from another speaker. As a consequence, the speech of the interlocutors may become more similar or less similar to each other. The former behavior is called convergence and the latter divergence. This phenomenon has been documented for human spoken interaction (e.g., Levitan & Hirschberg, 2011; Lewandowski, 2012; Pardo, 2006) and contributes to its success and quality (e.g., Borrie et al., 2015; Lee et al., 2010; Manson et al., 2013). Phonetic accommodation manifests at a holistic level—for example, when examining the long-term average spectrum of speech (Gregory & Webster, 1996)—down to the sub-phonemic level—for example, when measuring the voice onset time of plosives (e.g., Nielsen, 2011). Among the assumed reasons for this phenomenon are both an automatic alignment process (cf. Interactive Alignment Model; Pickering & Garrod, 2004, 2013, 2021) and a socially induced motivation (cf. Communication Accommodation Theory; Giles, 2016; Giles et al., 1991; Shepard et al., 2001). Phonetic accommodation has also been documented for human–computer interaction with humans adapting to the speech output of virtual interlocutors (e.g., Bell et al., 2003; Cohn et al., 2021; Gijssels et al., 2016; Suzuki & Katagiri, 2007; Zellou et al., 2021).

It is commonly observed that different speakers exhibit different degrees of phonetic accommodation (e.g., Pardo et al., 2018; Weise et al., 2019). Exploring the individual differences between speakers causing this variation is becoming a central point of accommodation research (e.g., Aguilar et al., 2016; Lewandowski & Jilka, 2019; Wade et al., 2021; Yu et al., 2013). One factor that may contribute to the individual differences in accommodating behavior is the general speaker disposition, which includes aspects such as innate phonetic talent, personality traits, and cognitive abilities. Only a few studies have investigated these aspects to date (Aguilar et al., 2016; Lewandowski & Jilka, 2019; Yu et al., 2013). However, their results suggest that it is promising to further examine the influence of speaker disposition on phonetic accommodation.

The present study thus examines the concept of mental boundaries as a specific component of speaker disposition, which has not been previously considered in accommodation research. The concept of mental boundaries was introduced as a dimension of human personality and an aspect of the overall organization of the mind by Ernest Hartmann and colleagues (Hartmann, 1989, 1991; Hartmann et al., 1987). To measure the strength of these mental boundaries, they developed the Boundary Questionnaire (BQ), which made the concept of (ego) boundaries, often described in the psychological literature (e.g., Fisher & Cleveland, 1968; Landis, 1970; Lewin, 1935), empirically explorable.

Our goal is to explore the relationship between the psychological concept of mental boundary strength and the linguistic phenomenon of phonetic accommodation. To enable the assessment of this personality factor in our accommodation research with German native speakers, we develop a German adaptation of a BQ short version.

To situate our work in the context of previous research, we first provide an introduction to the concept of mental boundaries (see Mental boundaries); motivate a possible relationship between mental boundaries and phonetic accommodation (see Boundaries and phonetic accommodation); give an overview of the development and availability of instruments measuring boundary strength in English—the origin language of the concept—and German—the target language of our research (see Measuring boundary strength); and take into consideration the relationship between mental boundaries and the Big Five, because the latter is the only personality-related aspect that has been explored in connection with phonetic accommodation to date (see Boundaries and the Big Five). The final section of the introduction then provides the background and outline of the present study, which we conducted using a corpus of spoken human-computer interaction collected in Gessinger, Möbius, et al. (2021) (see Background and outline of the present study).

1.1 Mental boundaries

The concept of mental boundaries was developed by Ernest Hartmann and colleagues (Hartmann, 1989, 1991; Hartmann et al., 1987) to characterize people who suffer from nightmares because they could not be grouped under the umbrella of other personality traits, such as the Big Five (Costa & McCrae, 1992), for example. Mental boundaries refer to both the degree of interconnectedness within the mind and between the mind and the surrounding world. While thick boundaries stand for clear demarcation, thin boundaries represent permeability.

Hartmann (1991) emphasizes that the concept of boundaries in the mind is very broad and includes personality traits that one would not necessarily assume to be related. This is also illustrated by the different types of boundaries considered in the concept, for example: boundaries related to thoughts and feelings, sleep-dream-wake boundaries, boundaries related to memory, boundaries around oneself, boundaries in organizing one’s life, boundaries in decision making and action, and others.

A person with thicker mental boundaries has a clearer sense of separation between, for example, their self and their surroundings; dream and reality; past, present and future; their own versus another group, whereas a person with thinner boundaries tends to merge such concepts. Therefore, the thick-boundary type functions largely unaffected by the environment, while the thin-boundary type tends to absorb the full extent of sensory input (A. Harrison & Singer, 2013; Hartmann et al., 2001). The boundary construct describes the general nature of a person’s mental boundaries as a value-free dimension of their personality, that is, without favoring one type over the other. ¹

Apart from this, the concept can also describe intra-individual differences, that is, a person can behave in a more thin or thick boundary manner depending on the situation (Hartmann et al., 2001). Generally, it was found that women and younger people have thinner mental boundaries than men and older people (Hartmann, 1991; McCrae, 1994). Typical examples of individuals with thinner mental boundaries are arts and music students (Beal, 1998) as well as people suffering from nightmares (Cowen & Levin, 1995; Hartmann, 1991; Levin et al., 1991); thicker mental boundaries can be found, for instance, in salespeople and lawyers, as well as in people with a preference for concreteness and order (R. Harrison et al., 2006; Hartmann, 1991).

The main field of application of the boundary construct to date is dream research, which is also where it originated from (e.g., Aumann et al., 2012; Hartmann & Kunzendorf, 2006). However, we believe that mental boundaries could also have the potential to distinguish individuals with a higher and lower tendency to phonetically accommodate to a conversational partner.

1.2 Boundaries and phonetic accommodation

As mentioned above, Hartmann (1991) describes various types of boundaries that are encompassed by the boundary construct. Among these are perceptual boundaries as well as interpersonal boundaries, boundaries of identity, and group boundaries. These seem to be particularly relevant in the context of phonetic accommodation, because we can assume that perceptual receptivity is a prerequisite for the latter and that permeable interpersonal relationships and fluid personal and social identities (i.e., group membership) facilitate accommodation given its social function. For intergroup contexts in particular, the notion of boundary permeability is echoed in the Communication Accommodation Theory, a generalized model of communicative interaction (Giles, 2016; Giles et al., 1991; Shepard et al., 2001), where it is assumed that weak identification with a group and soft intergroup boundaries predispose a speaker to converge to an interlocutor who is part of a different group (e.g., Gallois et al., 2005).

Language learning, especially acquiring the pronunciation of a foreign language, which can be interpreted—at least to some degree—as a manifestation of phonetic accommodation, may again serve as an example here. Hartmann (1991, p. 221) proposed en passant that thinner mental boundaries would be advantageous in learning “to speak a foreign language as native speakers do.” Already prior to that (e.g., Guiora et al., 1972), and then later with reference to Hartmann’s concept (e.g., Baran-Lucarz, 2012; Więckowska, 2011), authors suggested the permeability of ego boundaries as a facilitator for achieving native-like pronunciation in a foreign language.

According to Guiora (1994), pronunciation is anchored in the speaker’s identity. It is central to the language ego, which encompasses all components of self-representation through language and is surrounded by ego boundaries that are more or less flexible. These boundaries, and thus the self-representation, must be temporarily softened to speak a foreign language.

The few studies that have been conducted in this context showed only small effects of enhanced pronunciation skills for thinner boundaries and in some cases had limitations, for example, the indirect assessment of the boundary strength (e.g., Guiora et al., 1972) or an unvalidated test for the pronunciation attainment (e.g., Baran-Lucarz, 2012).

It has been suggested that language learners prefer different approaches depending on the nature of their mental boundaries. According to Ehrman (1999), learners with thick mental boundaries prefer conscious, linear processes like formal explanation and drilling. Thin-boundary learners, on the other hand, prefer strategies of receptivity to outside influence and “make use of [. . .] native speakers as models with which to identify” (p. 68). They tend to enjoy unexpected learning events.

Another factor in language learning that is associated with boundary permeability is tolerance of ambiguity, ² which consists of the following three stages: (1) permitting information to access one’s mind, (2) dealing with contradictions and incomplete information in the mental system, and finally (3) integrating the new information with existing schemata, which is referred to as accommodation (Ehrman, 1996). Especially for the first two stages, thinner mental boundaries are considered beneficial, provided that these learners do not become overwhelmed by the abundance of incoming information.

Overall, these considerations and findings suggest that thin, permeable boundaries may be more likely to allow phonetic accommodation than thick, inflexible ones.

1.3 Measuring boundary strength

The strength of mental boundaries can be measured using a questionnaire. To make concepts such as the boundary construct more accessible and easier to use for researchers, it is desirable to develop short versions of the often extensive original questionnaires. Moreover, for the wider dissemination of an instrument, it is required to adapt the questionnaire for other languages, which means translating it, adapting it to the cultural specifics of the target population if necessary, and validating the resulting questionnaire.

In the following, we give an overview of the development and availability of instruments measuring boundary strength in English—the origin language of the concept—and German—the target language of our research.

1.3.1 The Boundary Questionnaire

The original Boundary Questionnaire (BQ; Hartmann, 1989, 1991; Hartmann et al., 1987) was developed in English and contains 145 items that operationalize the concept of boundaries at many different levels of behavior and experience, ranging from more abstract interpretations—for example, “My thoughts blend into one another”—to more literal interpretations—for example, “I like heavy solid clothing.” Respondents express their attitude toward items like “I feel unsure of who I am at times” or “I think children need strict discipline” on a 5-point scale from 0 (not at all true of me) to 4 (very true of me). To control for the tendency of respondents to agree with statements, 58 of the 145 items, including the second example above, are reverse-scored. The overall boundary score SumBound is determined by summing the points; higher total numbers represent thinner boundaries. Seven items were excluded from the calculation of SumBound due to non-significant or negative correlations with the latter. However, they remained in the questionnaire. SumBound_BQ is therefore based on 138 items.

For a sample of 866 participants (53% female, 47% male; M_age = 33 years) who took the BQ, Hartmann (1991) reports a very high internal consistency (Cronbach’s $\alpha$ = .93), which suggests that some items may be redundant (Tsang et al., 2017) and the questionnaire could be shortened.

The validity of the BQ was assessed by examining subjects for whom particularly thin (art students and nightmare sufferers) or particularly thick (naval officers) mental boundaries were assumed according to theory. These predictions were confirmed in the data, with the thin group scoring higher SumBound values than the thick group.

1.3.2 Short versions of the BQ

There are two English short versions of the BQ, which could be used as a basis for an adaptation to German. In the following, we present both versions and explain our choice.

Kunzendorf et al. (1997) propose an 18-item version of the BQ (BQ18) for which they selected 18 of the 138 original items according to the following three criteria: the items’ face validity (i.e., their apparent relevance to the context of the questionnaire), their correlation with BQ’s SumBound, and their distribution over BQ’s topical categories. They refer to an unpublished study with 856 participants that showed a substantial correlation of SumBound_BQ18 with SumBound_BQ ( $r$ = .87). Thus, the 18-item version of the BQ seems to capture the boundary construct very well.

Rawlings (2001) developed an empirically-derived short version of the Boundary Questionnaire (BQ-Sh) ³ from the data of 300 participants (74% female, 26% male; M_age = 19 years) who completed the original BQ. Factor extraction was carried out using maximum likelihood factor analysis (MLFA) and subsequent Promax transformation of seven factors in a first step. In a second step, the size of the questionnaire was reduced by removing items with very low loadings or high cross-loadings, as well based on reliability analyses of the subscales derived in the first step. Finally, a further factor analysis resulted in six factors that were interpreted as: I: Unusual Experiences (UE), II: Need for Order (NfO^r), III: Trust (Tr), IV: Perceived Competence (PC^r), V: Childlikeness (Ch), and VI: Sensitivity (Se). ⁴

Although Factor VI included only two items, Rawlings (2001) decided to retain this subscale because “it was shown to have reasonable ‘reliability,’ to be conceptually meaningful, and to represent aspects of the construct not covered by the other factors” (pp. 135). In contrast, Factor III remained in the BQ-Sh, but was not included in the calculation of SumBound for conceptual reasons. Overall, there were no substantial correlations between the five subscales included in the calculation of SumBound_BQ-Sh (I, II, IV–VI), with $\left|r\right|<.3$ in all cases. According to Rawlings (2001), this demonstrates the complexity of the boundary construct and the fact that individuals do not necessarily have weak boundaries with respect to all aspects identified.

In terms of reliability, Rawlings (2001) reports adequate internal consistency for SumBound_BQ-Sh ( $\alpha$ = .74), which shows that shortening the questionnaire reduced item redundancy to an appropriate level. In addition, SumBound_BQ-Sh correlated substantially with SumBound_BQ ( $r$ = .88), demonstrating that the short version captures the boundary construct very well.

While the BQ18 clearly stands out for its brevity, the BQ-Sh with its effectively 40 items—when excluding the six items of the Tr subscale—is a very concise short version too. Only eight items appear in both short versions. In the sample of Rawlings (2001) the internal consistency was lower for the BQ18 ( $\alpha$ = .66) than for the BQ-Sh ( $\alpha$ = .74) and the two short versions correlated with $r$ = .77. As the BQ-Sh is based on a solid empirical foundation and introduces five fairly independent subscales that could contribute to further insights into the concept of mental boundaries, we used the BQ-Sh as a basis for the German adaptation.

1.3.3 German versions of the BQ

There are a number of authors in the field of dream research who refer to a German translation of the BQ long version (e.g., Funkhouser et al., 2001; Schredl, 2004; Strauch & Meier, 1999). However, this translation has not been published and is thus not accessible. Moreover, there is no information available about the adaptation process.

To our knowledge, there is no German adaptation available for the BQ-Sh. ⁵ We therefore believe that a systematic German adaptation of the BQ-Sh, consisting of the translation and validation of the instrument (see Questionnaire adaptation), can offer an important contribution to future research on mental boundaries.

1.4 Boundaries and the Big Five

The boundary concept is to some extent related to the Big Five personality traits Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. For instance, McCrae (1994) compared the BQ results of 124 subjects (57% female, 44% male; age range: 26–91 years) with results in the NEO Personality Inventory (NEO-PI; McCrae & Costa, 1985), which assesses the Big Five, and found significant correlations of SumBound with Neuroticism ( $r$ = .32), Extraversion ( $r$ = .27), and Openness ( $r$ = .66). Furthermore, in a joint factor analysis of various personality and cognitive ability measures for a subset of 85 respondents resulting in six factors reflecting the Big Five and general intelligence, SumBound_BQ had high loadings on the factors representing Neuroticism (0.44) and Openness (0.63). This indicates a connection between thin mental boundaries and these personality dimensions, which is promising with regard to our assumption that thinner boundaries are more likely to enable phonetic accommodation, as the few findings available in this context so far suggest that Openness and Neuroticism are predictive of such adaption. Yu et al. (2013) investigated word-initial voice onset time in a non-conversational phonetic imitation task in English and found that Openness was positively correlated with the degree of voice onset time convergence. Lewandowski and Jilka (2019) examined word-based amplitude envelope match in conversational interactions between native and non-native speakers of English. They found that speakers with higher levels of Neuroticism and Openness showed more convergence.

Hartmann et al. (2001) refer to the remarkably high correlation of SumBound and the NEO-PI factor Openness to Experience. They observe that the Openness concept of the NEO-PI assesses only such qualities of Openness that are perceived positively, for example, enjoying abstract ideas and speculations about the nature of the universe. In contrast, the Openness concept of the BQ also considers the more negative aspects of Openness, such as feeling overwhelmed or vulnerable, or becoming over-involved. They therefore consider the boundary construct as a more comprehensive approach to Openness that extends beyond the concept covered by the Big Five.

Schredl (2004) examined the relationship between the Big Five, assessed with the German version of the Revised NEO Personality Inventory (Ostendorf & Angleitner, 2004), and SumBound, assessed with the abovementioned unpublished German translation of the BQ, in a sample of 444 subjects (see German versions of the BQ) and found significant correlations for Neuroticism ( $r$ = .33), Openness ( $r$ = .54), Agreeableness ( $r$ = .18), and Conscientiousness ( $r$ = −0.35). ⁶

We included the Big Five in prior and present analyses to investigate whether the relationships reported above replicate in our data and to compare their predictive potential for phonetic accommodation with that of mental boundary strength.

1.5 Background and outline of the present study

The present study revisits data gathered in an experiment investigating the phonetic accommodation of human speakers to Mirabella, a virtual language learning tutor for German (see Mirabella corpus). A detailed description of the experiment and prior results can be found in Gessinger (2022) and Gessinger, Möbius, et al. (2021). Crucially, we demonstrated in this earlier work that the study participants converged to Mirabella’s speech output with respect to the realization of the word ending 〈-ig〉 and the intonation of wh-questions in German. We also explored whether the participants’ Big Five personality traits predict this accommodation (more details below). The present study extends the analysis of the Mirabella corpus and examines the extent to which accommodation can be predicted by the participants’ mental boundary strength. To this aim, we first developed a German version of the BQ-Sh (see Questionnaire adaptation), which the participants of the accommodation study then completed.

We hypothesize that thinner mental boundaries favor accommodation to an interlocutor because they stand for permeability, place less emphasis on group membership and separation from others, and allow for the absorption of external input (A. Harrison & Singer, 2013; Hartmann et al., 2001; see Boundaries and phonetic accommodation). This assumption is supported by the fact that a correlation of thin boundaries and the Big Five personality traits Neuroticism and Openness has been demonstrated (Hartmann et al., 2001; McCrae, 1994), and the same personality traits have also been found to facilitate phonetic accommodation—in the few studies conducted on this subject (Lewandowski & Jilka, 2019; Yu et al., 2013; see Boundaries and the Big Five).

The prior analyses of the German Mirabella corpus in Gessinger (2022) and Gessinger, Möbius, et al. (2021) also tested the influence of the Big Five and found that speakers with higher values for Neuroticism were more likely to converge to their interlocutor’s question intonation and a tendency ⁷ for speakers with higher values for Conscientiousness to be more likely to adopt the pronunciation variant of the word ending 〈-ig〉 from their interlocutor. The effect of Neuroticism therefore appears to be more robust than that of Conscientiousness. Regarding the question as to why the two personality traits would influence different phonetic features, we speculated that this could be due to the different reasons for accommodation that these personality traits entail. For example, speakers with a high degree of Neuroticism may accommodate to avoid communicative distress (Borkenau & Ostendorf, 2007; Lewandowski & Jilka, 2019), while speakers with a high degree of Conscientiousness may accommodate to perform well in the interaction (Borkenau & Ostendorf, 2007). These different reasons may promote convergence with respect to different types of phonetic features. For example, performing well may translate to using “correct” segments, while avoiding communicative distress may translate to producing questions with the same intonation as an interlocutor (Gessinger, 2022).

While the influence of Conscientiousness on accommodation reported in Gessinger (2022) is only a tendency, it is striking that Schredl (2004) reported a negative relationship between boundary strength and Conscientiousness for a German participant sample (see Boundaries and the Big Five). Should this relationship replicate in the present data, we would expect thicker mental boundaries to be related with higher levels of Conscientiousness—and hence with more accommodation for the word ending 〈-ig〉.

As the BQ-Sh, which we adapted for German in this study, consists of five fairly independent subscales that capture different aspects of mental boundaries (see Short versions of the BQ), we further investigate the predictive potential of the individual subscales for phonetic accommodation. This may allow us to identify aspects of boundary strength that are particularly conducive to accommodative behavior.

2 Material and methods

2.1 Questionnaire adaptation

The adaptation process consisted of translating the BQ-Sh to German and validating the resulting BQ-Sh-G with native speakers of German. The Big Five were used to assess the validity of the adaptation, because they had previously been considered in research on both mental boundaries and phonetic accommodation. The adaptation of the BQ-Sh for German was prepared taking the ITC Guidelines for Translating and Adapting Tests (International Test Commission, 2018) into consideration. The validation follows the method applied by Rawlings (2001) when developing the BQ-Sh. A full validation of the adaptation was beyond the scope of this interdisciplinary project and is strongly recommended as future work.

2.1.1 Translation

We used a double forward translation and reconciliation procedure in which the first author of this article and a professional translator independently translated the 46 items of the BQ-Sh from English into German and then discussed and reconciled any discrepancies between the two translations. The result was assessed by the second author of this article. All three translators involved in the process were native speakers of German and familiar with the target culture, namely German-speaking regions in Central Europe.

The result of this process is the BQ-Sh-G, which contains 46 items. Like the BQ-Sh, we kept the Trust (Tr) subscale in the questionnaire for use in future research, although it is not included in the calculation of SumBound. The BQ-Sh-G can thus also be administered in a more compact 40-item version that does not include the Tr subscale. See Supplementary Material (A) for the full BQ-Sh-G and (B) for further details about the translation process.

2.1.2 Validation

The validation sample consisted of 335 native speakers of German (64% female; 36% male; age M = 34.11 years, SD = 12.83, range: 18–79). See Supplementary Material (C) for further details about the sample. Participants completed the BQ-Sh-G and the NEO-FFI (Borkenau & Ostendorf, 2007).

Figure 1 gives an overview of how the total SumBound score is distributed among the validation sample. Theoretically achievable are scores from 0 to 160 (40 items × 0 to 4 points). The SumBound scores in the total sample range from 39 to 118, with 50% of them falling between 65 (Q1) and 82 (Q3). Mean SumBound of the BQ-Sh-G in the validation sample is 73.61 (SD = 12.91). See Supplementary Material (D) for details about the effect of gender and age on mental boundary strength.

Figure 1.

Distribution of the BQ-Sh-G total score (SumBound) in the validation sample $(n$ = 335). The lines indicate the 25% (Q1) and 75% (Q3) quantiles.

To explore the internal structure of the BQ-Sh-G data, we conducted a maximum likelihood factor analysis (MLFA). Details of this analysis can be found in Supplementary Material (E).

We assessed the reliability of the BQ-Sh-G using Cronbach’s alpha coefficients. SumBound ( $\alpha$ = .77) and most subscales (UE: 0.79, NfO^r: 0.79, Tr: 0.71, and Ch: 0.69) show adequate internal consistency with Cronbach’s alpha close to or above 0.7; only the coefficients of the PC^r and Se subscales are somewhat lower with $\alpha$ = .65 and .64, respectively. Recall, however, that the Se subscale contains only two items. Supplementary Material (F) provides further descriptive statistics and all reliability coefficients for both the German BQ-Sh-G and the English BQ-Sh. We also show correlation matrices of total scores and their subscales for both instruments in Supplementary Material (G).

The internal consistency of the five NEO-FFI subscales in our data is adequate, with alpha coefficients of 0.88 for Neuroticism, 0.82 for Extraversion, 0.7 for Openness to Experience, 0.76 for Agreeableness, and 0.85 for Conscientiousness.

Figure 2 shows a correlation matrix of the BQ-Sh-G and the NEO-FFI data. The Neuroticism and Openness to Experience subscales of the NEO-FFI correlate positively with the BQ-Sh-G SumBound score: 0.32 and 0.35, respectively; the Conscientiousness subscale shows a negative correlation of −0.47 with SumBound. Some moderate to strong correlations between the NEO-FFI and BQ-Sh-G subscales emerge as well: Neuroticism with UE (0.41) and Se (0.52); Extraversion with Tr (0.45); Conscientiousness with PC^r (−0.66). There are also several weak correlations: Extraversion with Ch (0.29); Openness to Experience with Tr (0.2), NfO^r (0.29), and UE (0.21); Agreeableness with Tr (0.26); Conscientiousness with NfO^r (−0.29). ⁸

Figure 2.

Correlation matrix of the BQ-Sh-G total score (SumBound) and subscales Unusual Experiences (UE), Need for Order (NfO^r), Trust (Tr), Perceived Competence (PC^r), Childlikeness (Ch), Sensitivity (Se) with the NEO-FFI subscales Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness. Given correlations are significant with $p<$ .05 or smaller, Bonferroni-corrected for the total number of comparisons ( $n$ = 35).

2.1.3 BQ-Sh-G reliability

Overall, the total SumBound score and the subscales of the German BQ-Sh-G exhibit means and standard deviations similar to those of the English BQ-Sh in Rawlings (2001)—see Supplementary Material (F). As reported above, they also show adequate internal consistency. It is noteworthy that the data in Rawlings (2001) came from early-stage psychology students, whereas the present data were collected from a sample with a wide range of ages and professional backgrounds—see Supplementary Material (C). This makes the current sample more representative of the general population.

All subscales of the BQ-Sh-G had significant positive correlations with SumBound—see Supplementary Material (G). This warrants the use of the sum score for assessing boundary strength—as intended by the original author of the concept, Ernest Hartmann. In summary, we demonstrated that the German adaptation can be used equivalently to the BQ-Sh. Future work may address further validation and refinement of the instrument.

2.1.4 BQ-Sh-G and the Big Five

In terms of convergent validity, there seems to be a relationship between boundary strength and the Big Five personality dimensions, which manifests itself repeatedly: Thinner boundaries tend to be associated with increased Neuroticism and Openness to Experience (cf. McCrae (1994), Schredl (2004), and the present data). This finding supports our hypothesis that thinner boundaries facilitate phonetic accommodation, because previous evidence from the literature suggests that Openness and Neuroticism favor such accommodation (Lewandowski & Jilka, 2019; Yu et al., 2013). Regarding the relationship with Conscientiousness, the situation is more ambiguous. To our knowledge, a negative correlation of boundary strength with the latter has so far only been found for a German-speaking population (Schredl, 2004)—and also occurred in the present data. We see no reason why English- and German-speaking target populations should behave differently with respect to this relationship. Future work may establish whether this finding holds.

The observed correlations of SumBound with the subscales of the NEO-FFI are only weak to moderate, which demonstrates that the BQ-Sh-G possesses discriminant validity, too. This is also reflected in the comment by Hartmann et al. (2001) on the relationship between SumBound and Openness to Experience, mentioned above, stating that the Openness concept of the BQ encompasses negative aspects as well and is thus a more comprehensive approach to Openness that goes beyond the concept captured by the Big Five.

Overall, we were able to demonstrate convergent and discriminant validity between our German adaptation of the BQ-Sh and the Big Five to a satisfactory degree. ⁹

2.2 Mirabella corpus

The Mirabella corpus contains speech data from 42 native speakers of German collected in a laboratory experiment (Gessinger, Möbius, et al., 2021). Recordings were made in a sound-attenuated booth with a stationary cardioid microphone at a sampling rate of 48 kHz. The experiment was framed as computer-assisted language learning (CALL) to motivate the exchange with Mirabella, a female virtual language learning tutor for German, for the participants. Note that we are reporting the data of German native speakers, who were instructed to “test” the CALL application. While Mirabella was presented as an autonomous system, she was actually controlled by the experimenter in a Wizard-of-Oz setup (Dahlbäck et al., 1993). Participants spoke to one of two Mirabella versions, the first used natural speech recorded by a female native speaker of German, the second used synthetic speech generated with MaryTTS (Le Maguer et al., 2018), the Hidden Markov model-based Speech Synthesis System (HTS; Zen & Toda, 2005), and WORLD (Morise et al., 2016). Phonetic accommodation was observed to a similar degree for both versions of Mirabella. Both conditions are therefore jointly assessed in the present analysis. In the experiment, the participants communicated with Mirabella through speech and solved a series of tasks presented on a computer screen. The entire interaction took about 30 min. Participants were found to converge to Mirabella’s speech output regarding the following two tested features: (1) the German word ending 〈-ig〉 (e.g., in “König,” engl. king) being produced with a final fricative (i.e., [ɪç]) or a final plosive (i.e., [ɪk]), and (2) the intonation of wh-questions (e.g., “Wo hat sich der Affe versteckt?,” engl. Where did the monkey hide?) being overall falling or rising. ¹⁰

2.2.1 Participants

The 42 native speakers of German who participated in the experiment with Mirabella (74% female, 26% male; age M = 24.57 years, SD = 6.73, range: 18–55) were recruited from Saarland University in Saarbrücken, Germany, and compensated for their participation. They are a subset of the 335 participants whose data were used in the development of the BQ-Sh-G. The BQ-Sh-G and NEO-FFI were administered after their participation in the experiment with Mirabella. Mental boundary strength was determined by calculating the SumBound score for each participant. Figure 3 gives an overview of how the total SumBound score is distributed among the participants. It ranges from 49 to 118, with 50% of the scores falling between 71.25 (Q1) and 82.75 (Q3). Mean SumBound of the BQ-Sh-G in this sample is 75.95 (SD = 12.82).

Figure 3.

Distribution of the BQ-Sh-G total score (SumBound) in the participant sample $(n$ = 42). The lines indicate the 25% (Q1) and 75% (Q3) quantiles.

2.2.2 Tasks and features

The participants solved a series of tasks presented on a computer screen in interaction with Mirabella, the means of communication being speech. The first investigated feature was the realization of the word ending 〈-ig〉. Native speakers of German use [ɪç] and [ɪk] interchangeably, but usually have a clear preference for one variant. The frequency of occurrence varies regionally, yet this variation is not perceived as a strong dialectal marker (Kiesewalter, 2019; Kleiner, 2011). The feature was tested in two separate tasks: In an initial picture naming and adjective translating (English to German) task, during which Mirabella did not speak yet, the experimenter, a trained phonetician, determined the preferred variant (majority out of 12 item productions) of each participant in real-time by auditorily categorizing productions as [ɪç] or [ɪk]. Any fricative variants (e.g., [ɪʃ] or [ɪɕ]) were categorized as belonging to the [ɪç] class. In the rare event of a tie, the Standard German variant [ɪç] was set as the speaker preference. Mirabella (i.e., the experimenter) accepted allophonic variation to avoid participants changing their pronunciation because they were not understood, but she did not accept synonyms of target items to retain the relatively inflexible character of a computer system. Even though some speakers used both variants interchangeably as expected, a preference could be established for each participant in this way. The preference was [ɪç] for 17 participants and [ɪk] for 25 participants, with 69% producing their preferred variant 10 to 12 times out of 12, 29% producing their preferred variant 7 to 9 times, and only one participant producing both preferred and dispreferred variants 6 times each (thus assigned the preference [ɪç]).

In a map task following later in the interaction, Mirabella used the respective dispreferred variant in her own speech output when providing the participants with missing information needed to complete the task. She retained this variant throughout the entire interaction and did not adapt to the participants. The participants had to describe a path from a starting point to a destination that led past various object-adjective pairs. In total, there were four maps with nine objects, three of which were 〈-ig〉 targets in each case.

After the experiment, two phonetically trained annotators independently classified all realizations of 〈-ig〉 from both tasks as belonging to the fricative or plosive class. The annotators agreed in all cases. Only a single 〈-ig〉 instance was misclassified during the real-time annotation of the first author during the experiment, which demonstrates that the participant preference was determined correctly.

The picture naming and adjective translating task served as baseline condition and the map task as testing condition for convergence. Accommodation in the form of convergence was demonstrated by an increase in the number of dispreferred variants, that is, the same variant that Mirabella produces, in the testing condition. ¹¹

The second phonetic feature under investigation, the intonation of wh-questions, was tested in a question-and-answer game of two rounds. The participants and Mirabella took turns in asking and answering questions about the location of 10 different animals displayed on the screen. Mirabella produced wh-questions with overall falling intonation in the first round and overall rising intonation in the second round. The default expectation is for wh-questions in German to be produced with falling contours. Rising contours are mainly applied when a question is uttered again, because the answer was not understood (Grice & Baumann, 2002). As the same animals were shown in both rounds of the game (information already provided in round 1), but their location on the screen differed between the rounds (new information to be provided in round 2), it was pragmatically correct, yet not expected that the participants would produce rising contours in the second round of the game—unless they converged to Mirabella’s speech output.

After the experiment, two trained phoneticians independently classified all intonation contours and found falling, falling-rising, and rising contour types in the data (Grice & Baumann, 2002), with the latter two types being treated as overall rising contours in the subsequent analysis. The annotation also took the position of the nuclear pitch accent into account, which is beyond the scope of the present paper (see Gessinger, Möbius, et al. (2021) for details).

The first round of the question-and-answer game served as baseline condition and the second round as testing condition for convergence. Accommodation in the form of convergence was demonstrated by an increase in the number of rising contours in the testing condition. ¹²

2.3 Statistical analysis

The two binary dependent variables discussed above—dispreferred versus preferred 〈-ig〉 variants and rising versus falling intonation contours—were analyzed using generalized linear mixed-effects models (GLMMs) formulated with the lme4 package (Bates et al., 2015) and evaluated with the lmerTest (Kuznetsova et al., 2017) and ggeffects (Lüdecke, 2018) packages in RStudio (Posit Team, 2024) with R (R Core Team, 2024).

Phonetic accommodation in the form of convergence is demonstrated by an increase in dispreferred 〈-ig〉 variants and an increase in rising F0 contours, that is, a main effect of the experimental condition (baseline vs. testing), as has been shown in Gessinger, Möbius, et al. (2021). The influence of boundary strength on phonetic accommodation would now be reflected in an interaction of SumBound with the experimental condition, indicating that boundary strength moderates the respective increase. The same would be the case for the influence of the individual BQ-Sh-G subscales. SumBound and the components contained in it, that is, scores for Unusual Experiences, Need for Order, Perceived Competence, Childlikeness, and Sensitivity, were tested in separate statistical models.

Model selection was carried out bottom-up, starting with a model which only included the random factor intercepts for PARTICIPANT and ITEM. Then, the binary factor CONDITION (sum coded levels: baseline/testing) and its interaction with the continuous factor BOUNDARY STRENGTH (SumBound; or one of the five subscale scores contained in SumBound) were added to the model. Random slopes for CONDITION by PARTICIPANT were included as well. Random intercepts/slopes would only be removed to simplify the model in cases of convergence errors or to allow a non-singular fit. The influence on the model fit was assessed by means of the Akaike information criterion (AIC), which estimates the relative quality of a statistical model for a given data set by taking into account the likelihood function and the number of estimated parameters (Akaike, 1973). A factor was kept in the model if the model fit improved significantly, as determined by a likelihood-ratio test, and the AIC value decreased by at least two points as compared to the model without the factor in question. Factors kept in the model are being considered significant predictors of the respective dependent variable at $\alpha$ = 0.05. To account for the fact that SumBound and the five subscale scores contained in SumBound were tested separately, which increases the probability of a false positive result, adjusted p-values using the Benjamini–Hochberg procedure (Benjamini & Hochberg, 1995) are reported as well.

3 Results

Figure 4 shows the percentage increase in dispreferred 〈-ig〉 variants and rising F0 contours from baseline to testing condition for each participant. For both phonetic features, the increase is mostly positive, indicating convergence to Mirabella. There are several participants who maintain their way of speaking in the testing condition, that is, no increase is observed. Furthermore, a few participants show a decrease in dispreferred 〈-ig〉 variants or rising F0 contours, indicating divergence from Mirabella. Participants are also sorted by increasing SumBound values. For 〈-ig〉, high increases in the target realization appear to be slightly skewed toward lower values of SumBound, whereas for the F0 contour, high increases in the target intonation are distributed across the entire SumBound range.

Figure 4.

Percentage increase in dispreferred 〈-ig〉 variants (top) and rising F0 contours (bottom) from baseline to testing condition for each participant. The participants are sorted by increasing SumBound value, which is shown above each bar.

Figure 5 visualizes the influence of experimental condition and overall boundary strength on the examined phonetic features as predicted by the following two GLMMs for the 〈-ig〉 variant (1) and the F0 contour (2):

$$\begin{array}{l}\mathrm{i}\mathrm{g}\mathrm{V}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{t}\sim \mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}*\mathrm{B}\mathrm{O}\mathrm{U}\mathrm{N}\mathrm{D}\mathrm{A}\mathrm{R}\mathrm{Y}\mathrm{S}\mathrm{T}\mathrm{R}\mathrm{E}\mathrm{N}\mathrm{G}\mathrm{T}\mathrm{H}+\hfill \\ \left(1+\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}|\mathrm{P}\mathrm{A}\mathrm{R}\mathrm{T}\mathrm{I}\mathrm{C}\mathrm{I}\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{T}\right)+\left(1|\mathrm{I}\mathrm{T}\mathrm{E}\mathrm{M}\right)\hfill \end{array}$$

(1)

$$\begin{array}{l}\mathrm{f}0\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{o}\mathrm{u}\mathrm{r}\sim \mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}*\mathrm{B}\mathrm{O}\mathrm{U}\mathrm{N}\mathrm{D}\mathrm{A}\mathrm{R}\mathrm{Y}\mathrm{S}\mathrm{T}\mathrm{R}\mathrm{E}\mathrm{N}\mathrm{G}\mathrm{T}\mathrm{H}+\hfill \\ \left(1+\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}|\mathrm{P}\mathrm{A}\mathrm{R}\mathrm{T}\mathrm{I}\mathrm{C}\mathrm{I}\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{T}\right)\hfill \end{array}$$

(2)

For the [ɪç]/[ɪk] contrast, including CONDITION into the intercepts-only model statistically significantly improved the fit, with ${\chi }^2=196.32,$ $df=3,$ $p<.001$ and an AIC value decrease ( ${\Delta }_{AIC}$ ∆_AIC) of $190.32$ points. Adding the interaction with BOUNDARY STRENGTH further improved the model fit, ${\chi }^2=7.77,$ $df=2,$ $p=.021,$ ${\Delta }_{AIC}=3.77.$ The final 〈-ig〉 variant model is hence the one indicated in (1) above, which is illustrated by Figure 5(a) and (b). The illustration suggests that participants with thicker mental boundaries may be more likely to converge to Mirabella’s realization of 〈-ig〉. The predicted percentage of dispreferred variants differs between baseline and testing conditions overall, but more clearly so for lower SumBound scores—depicted as five-number summary (i.e., minimum, first to third quartile, and maximum) in 5(a) and as continuous measure in 5(b). However, the interaction of CONDITION and BOUNDARY STRENGTH did not reach statistical significance in the model, Estimate (log-odds) = $0.03,$ $SE=0.02,$ $z=1.92,$ $p=.056,$ $p_{adjust.}=.083.$ The model shows that CONDITION had a statistically significant effect on the amount of dispreferred 〈-ig〉 variants, Estimate (log-odds) = $-3.07,$ $SE=1.16,$ $z=-2.66,$ $p=.008,$ $p_{adjust.}=.016,$ with more dispreferred variants being produced in the testing condition. It also shows a statistically significant effect of BOUNDARY STRENGTH, Estimate (log-odds) = $-0.06,$ $SE=0.02,$ $z=-2.55,$ $p=.011,$ $p_{adjust.}=.029,$ indicating a higher occurrence of dispreferred variants in participants with thicker mental boundaries independently of the experimental condition, that is, more alternating use of [ɪç] and [ɪk] overall.

Figure 5.

Model predictions for the influence of experimental condition and boundary strength on the examined phonetic features. Figures (a) and (c) show percentages of dispreferred 〈-ig〉 variants and rising F0 contours predicted for the minimum value (49; thick), lower quartile (71), median quartile (77), upper quartile (83), and the maximum value (118; thin) of SumBound per experimental condition (baseline vs. testing). Figures (b) and (d) show continuous predictions for baseline condition (red) and testing condition (blue). Each figure displays the 95% confidence interval.

When considering the individual subscales of the BQ-Sh-G in separate statistical models, only including the interactions of CONDITION with UNUSUAL EXPERIENCES, ${\chi }^2=11.26,$ $df=2,$ $p=.004,$ ${\Delta }_{AIC}=7.2,$ and PERCEIVED COMPETENCE, ${\chi }^2=9.11,$ $df=2,$ $p=.011,$ ${\Delta }_{AIC}=5.11,$ improved the model fit. The following two GLMMs are hence illustrated in Figure 6:

$$\begin{array}{l}\mathrm{i}\mathrm{g}\mathrm{V}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{t}\sim \mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}*\mathrm{U}\mathrm{N}\mathrm{U}\mathrm{S}\mathrm{U}\mathrm{A}\mathrm{L}\mathrm{E}\mathrm{X}\mathrm{P}\mathrm{E}\mathrm{R}\mathrm{I}\mathrm{E}\mathrm{N}\mathrm{C}\mathrm{E}\mathrm{S}+\hfill \\ \left(1+\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}|\mathrm{P}\mathrm{A}\mathrm{R}\mathrm{T}\mathrm{I}\mathrm{C}\mathrm{I}\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{T}\right)+\left(1|\mathrm{I}\mathrm{T}\mathrm{E}\mathrm{M}\right)\hfill \end{array}$$

(3)

$$\begin{array}{l}\mathrm{i}\mathrm{g}\mathrm{V}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{t}\sim \mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}*\mathrm{P}\mathrm{E}\mathrm{R}\mathrm{C}\mathrm{E}\mathrm{I}\mathrm{V}\mathrm{E}\mathrm{D}\mathrm{C}\mathrm{O}\mathrm{M}\mathrm{P}\mathrm{E}\mathrm{T}\mathrm{E}\mathrm{N}\mathrm{C}\mathrm{E}+\hfill \\ \left(1+\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}|\mathrm{P}\mathrm{A}\mathrm{R}\mathrm{T}\mathrm{I}\mathrm{C}\mathrm{I}\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{T}\right)+\left(1|\mathrm{I}\mathrm{T}\mathrm{E}\mathrm{M}\right)\hfill \end{array}$$

(4)

Figure 6(a) and (b) suggest that fewer Unusual Experiences (UE) relate to a bigger increase in dispreferred 〈-ig〉 variants, and Figure 6(c) and (d) suggest that higher Perceived Competence (PC^r; as this scale is reverse-scored) relates to a bigger increase in dispreferred 〈-ig〉 variants. Both interactions were statistically significant in the respective model, for UE: Estimate (log-odds) = $0.08,$ $SE=0.03,$ $z=2.66,$ $p=.008,$ $p_{adjust.}=.042;$ for PC^r: Estimate (log-odds) = $0.11,$ $SE=0.04,$ $z=2.46,$ $p=.014,$ $p_{adjust.}=.042$ . Both models further show a main effect of condition, for UE: Estimate (log-odds) = $-1.92,$ $SE=0.43,$ $z=-4.49,$ $p<.001,$ $p_{adjust.}<.001;$ for PC^r: Estimate (log-odds) = $-2.78,$ $SE=0.79,$ $z=-3.53,$ $p<.001,$ $p_{adjust.}=.001,$ as well as a main effect of the respective BQ-Sh-G subscale, for UE: Estimate (log-odds) = $-0.11,$ $SE=0.04,$ $z=-2.73,$ $p=.006,$ , $p_{adjust.}=.029;$ for PC^r: Estimate (log-odds) = $-0.16,$ $SE=0.06,$ $z=2.44,$ $p=.015,$ $p_{adjust.}=.029.$ These main effects demonstrate phonetic accommodation in the form of convergence independently of the boundary strength components, as well as a higher number of dispreferred 〈-ig〉 variants in participants with fewer Unusual Experiences and higher Perceived Competence independently of the experimental condition.

Figure 6.

Equivalent to Figure 5, model predictions for the influence of the BQ-Sh-G subscales Unusual Experiences (a, b) and Perceived Competence (c, d) on accommodation to 〈-ig〉. Higher scores on the reverse-scored PC^r scale denote lower perceived competence.

For question intonation, including CONDITION into the intercepts-only model also statistically significantly improved the fit, with ${\chi }^2=426.83,$ $df=3,$ $p<0.001$ and an AIC value decrease (∆AIC ${\Delta }_{AIC}$ ) of $420.83$ points. However, adding the interaction with boundary strength did not further improve the model fit, ${\chi }^2=0.21,$ $df=2,$ $p=.9$ . ¹³ This is also illustrated by Figure 5(c) and (d). While the predicted percentage of rising contours differs between baseline and testing conditions overall, there is no substantial variation in this difference over the SumBound scores—depicted as five-number summary in 5(c), and as continuous measure in 5(d). The final intonation contour model is hence:

$$\begin{array}{l}\mathrm{f}0\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{o}\mathrm{u}\mathrm{r}\sim \mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}+\hfill \\ \left(1+\mathrm{C}\mathrm{O}\mathrm{N}\mathrm{D}\mathrm{I}\mathrm{T}\mathrm{I}\mathrm{O}\mathrm{N}|\mathrm{P}\mathrm{A}\mathrm{R}\mathrm{T}\mathrm{I}\mathrm{C}\mathrm{I}\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{T}\right)+\left(1|\mathrm{I}\mathrm{T}\mathrm{E}\mathrm{M}\right)\hfill \end{array}$$

(5)

It shows that CONDITION had a statistically significant effect on the amount of rising F0 contours, Estimate (log-odds) = $-4.15,$ $SE=1.04,$ $z=-4.00,$ $p<.001,$ with more rising contours being produced in the testing condition.

None of the individual subscales of the BQ-Sh-G significantly improved the model fit for question intonation when included in separate statistical models.

Table 1 provides a result overview of the present analysis and a comparison to the Big Five results in Gessinger (2022).

Table 1.

Overview of the results for the examined phonetic features, indicating whether the respective personality factor was included in the statistical model (based on significantly improved model fit) and whether the interaction with the experimental condition was statistically significant. The arrows illustrate whether a lower (↓) or higher (↑) score is associated with a higher degree of phonetic accommodation. As the perceived competence scale is reverse-scored, lower scores denote higher competence.

4 Discussion

We found some evidence for the prediction that thicker mental boundaries would favor convergence with respect to the binary [ɪç]/[ɪk] contrast for a group of German native speakers participating in an experiment involving the virtual language learning tutor Mirabella. Including the interaction of boundary strength with the task testing for phonetic accommodation in the statistical model yielded a better representation of the data, suggesting that the model captures additional variability when boundary strength is accounted for. However, the interaction was not statistically significant in the final model. That said, we identified a statistically significant effect of two BQ-Sh-G subscales: fewer Unusual Experiences (UE) and higher Perceived Competence (PC^r; reverse-scored) both related to a higher degree of 〈-ig〉 variant accommodation. We do not see a conceptual reason why these two SumBound components would be more likely to show an effect, while the remaining three components (Need for Order, Childlikeness, and Sensitivity) would be less likely. This inconsistency within the broader construct of boundary strength merits further exploration to better understand the underlying dynamics. Notably, UE and PC^r did not show a statistically significant correlation in the BQ-Sh-G validation sample (see Supplementary Material G). In line with Gessinger (2022), who found a tendency for the Big Five personality trait Conscientiousness to facilitate 〈-ig〉 variant accommodation, PC^r had a statistically significant strong negative correlation with Conscientiousness in the validation sample ( $-0.66$ ; see Figure 2). However, no such correlation with Conscientiousness was found for UE. Overall, it remains unclear why the UE and PC^r components of boundary strength stand out as predictors of accommodation in this data.

Recall that the specific prediction that thicker mental boundaries would facilitate accommodation for the [ɪç]/[ɪk] contrast was not consistent with our overarching intuition that thinner boundaries would facilitate convergence. Instead, this prediction emerged from our observations of the relationship between the Big Five and boundary strength on one hand, and accommodative behavior in our data on the other hand. That said, an inclination of individuals with thick mental boundaries to adopt a feature like [ɪç]/[ɪk] from their interlocutor, would be consistent with the fact that they value clear separation (A. Harrison & Singer, 2013; Hartmann et al., 2001)—for example, categorical nature of the contrast—and exhibit a high need for order (Rawlings, 2001)—for example, both interlocutors using the same variant. However, our finding that these participants show more alternation in the use of [ɪç]/[ɪk] independently of the interaction with Mirabella is not easily reconciled with the theoretical background of thick boundaries. Further studies would need to investigate whether this outcome replicates.

In keeping with the overarching intuition that thinner boundaries would promote convergence, we expected this to be the case for question intonation. In particular, perceptual receptivity, which we understand to be associated with thin perceptual boundaries (a boundary type encompassed in the boundary construct; Hartmann, 1991), seems relevant to prosodic adaptation. However, we found no evidence for thinner boundaries promoting the latter in our data. We observed only a slight tendency of the data in the predicted direction, which cannot be reasonably interpreted given the large variation.

We would like to acknowledge that the Mirabella corpus contains a relatively small number of participants. To increase the power of the analysis as much as possible given the data, we assessed the effect of boundary strength on phonetic accommodation jointly for participants who interacted with one of two Mirabella versions, using natural or synthetic speech. This was warranted because phonetic accommodation was observed to a similar degree for both Mirabella versions in a previous analysis and we have no theoretical reason to believe that the effect of boundary strength would interact with voice quality—although we cannot completely rule out this possibility. However, the effect of boundary strength on phonetic accommodation may still be too subtle to be robustly detected under these circumstances—especially considering that accommodation effects are usually subtle themselves (e.g., Gessinger, Raveh, et al., 2021; Pardo et al., 2017). To increase the probability of detecting a true effect, if it exists, and to strengthen the reliability of our results, it would be desirable to increase the number of participants in future studies.

In this context, we would also like to reiterate that the participants of the experiment with Mirabella were not selected based on their boundary composition. While we show that SumBound is spread across a wide range in our participants—49 to 118 of theoretically achievable 0 to 160—the majority fall in a medium range, as expected for the general population. Testing a group of participants with extreme expressions of mental boundaries (particularly thick/thin) would be a relevant extension for the present work. We assume that existing effects would be more evident in such a group. This would increase the power of the analysis, even if the number of participants remained relatively small (see approach of Aguilar et al., 2016).

In the case of the Mirabella corpus, we investigate phonetic accommodation between human and computer—or, more precisely, from the human toward a computer, that shows no phonetic accommodation itself. It remains an open question whether the effect of boundary strength might depend on the interlocutor type (computer vs. human) and interlocutor behavior (accommodation vs. no accommodation), or even on the perceived boundary strength of the interlocutor. As we understand mental boundary strength to be a fundamental disposition, we do not expect that a computer interlocutor would entirely prevent an effect if it were present for a human interlocutor. However, it has been shown that different feature constellations between interlocutors can influence the dynamics of phonetic accommodation (e.g., same-sex vs. mixed-sex partners, Levitan et al., 2012; partners matched vs. mismatched in terms of rejection sensitivity, Aguilar et al., 2016; partners differing in social status, Gregory & Webster, 1996). Therefore, future work should consider a variety of dialogue partner constellations when exploring the effect of boundary strength on phonetic accommodation in more detail.

In the case of mental boundaries, we are considering a very broad concept that combines many personality dimensions (Hartmann, 1991). While it covers Neuroticism and Openness at one end of the continuum and Conscientiousness at the other to some extent (McCrae, 1994; Schredl, 2004), it cannot be regarded as a substitute for the separate evaluation of the Big Five. In the Mirabella data, both constructs have shown some predictive power for convergence with respect to the phonetic features under investigation. The Big Five trait Neuroticism predicted F0 contour accommodation, and the two boundary strength subscales UE and PC^r predicted 〈-ig〉 variant accommodation.

The current study presents a first exploration of the relationship between mental boundary strength and phonetic accommodation. It focuses on very specific phonetic features. Similarly, the few studies that examined the influence of the Big Five on accommodation also investigated specific phonetic features, that is, word-based amplitude envelope match (Lewandowski & Jilka, 2019), word-initial voice onset time (Yu et al., 2013), and our own prior work on the Mirabella corpus (Gessinger, 2022; Gessinger, Möbius, et al., 2021).

On one hand, we believe that it is possible to identify feature types—for example, characterized by their nature (e.g., categorical vs. continuous) or their communicative function (e.g., marked vs. unmarked pronunciation variant)—that relate to the needs of individuals with different personality structures in spoken interaction. Speakers may (consciously or unconsciously) consider a feature that appears to be conducive to these respective needs as a particularly “worthwhile” target for accommodation. To test this hypothesis, a principled way of determining which features these might be would need to be established.

On the other hand, given its broad nature, the boundary construct does not lend itself intuitively to predictions about specific features, but rather to predictions about general tendencies. Boundary strength may be more predictive of a general disposition to accommodation, which may be eclipsed when individual phonetic phenomena are considered. For example, a speaker who does not converge to Mirabella’s way of producing 〈-ig〉, may still converge to a host of other phonetic features in her speech which may lead to the overall impression of convergence. Prior research on phonetic accommodation has indeed shown that performance of individual acoustic dimensions does not necessarily correspond to the degree of accommodation perceived by listeners, with perceptual measures integrating over various dimensions, that is, providing a holistic impression of phonetic accommodation (e.g., Pardo, 2013; Pardo et al., 2017). Ostrand and Chodroff (2021) approximated the holistic, perceptual level by jointly analyzing accommodation of more than 300 acoustic-phonetic features using a machine learning classifier. They demonstrated that the holistic measure sometimes aligns with the performance of individual phonetic features and sometimes does not.

As mentioned in the beginning, the literature suggests that thinner mental boundaries may be conducive to the acquisition of native-like pronunciation in a foreign language (e.g., Baran-Lucarz, 2012; Guiora et al., 1972; Więckowska, 2011). For this scenario, it also seems reasonable to consider a holistic view of native-like pronunciation, because performance with respect to an isolated phonetic feature certainly contributes to the overall impression, but does not necessarily determine it.

While we observe considerable variability in the accommodation behavior of the same speaker when comparing different phonetic features (e.g., Cohen Priva & Sanker, 2018; Gessinger, Raveh, et al., 2021; Sanker, 2015), we also know that there are speakers who have a greater tendency to converge to their conversational partner than others (e.g., Pardo et al., 2018; Weise et al., 2019). ¹⁴ This is where the broad concept of mental boundaries could complement the previously assumed influencing factors for phonetic accommodation. Future studies on accommodation behavior—especially those examining a large number of phonetic features simultaneously—should consider boundary strength as a possible influencing factor to shed further light on its predictive potential.

5 Conclusion

We motivated a possible relationship between mental boundaries and phonetic behavior in communicative interaction, proposing that thinner boundaries indicate a greater disposition to adapt to an interlocutor’s speech output. To explore this relationship, we made a German adaptation of the Boundary Questionnaire’s (BQ; Hartmann, 1991) empirically derived short version (BQ-Sh; Rawlings, 2001) available, that is, the BQ-Sh-G. We explored the relationship between the BQ-Sh-G and the Big Five personality traits, because they have previously been considered in research on both mental boundaries and phonetic accommodation. We found positive correlations of boundary strength with Neuroticism and Openness to Experience, as well as a negative correlation with Conscientiousness, which is in line with previous literature. We strongly encourage further validation of the BQ-Sh-G, which was beyond the scope of this interdisciplinary project.

We collected BQ-Sh-G scores from participants of an experiment on phonetic accommodation in a human–computer interaction context. In this experiment, native speakers of German interacted with the virtual language learning tutor Mirabella. We found a tendency for thicker mental boundaries to promote convergence to Mirabella’s pronunciation variant of the German word ending 〈-ig〉 as either [ɪç] or [ɪk], which was driven by statistically significant effects of the BQ-Sh-G subscales Unusual Experiences and Perceived Competence. We compared this result to prior results on the same corpus demonstrating that a higher degree of Neuroticism promoted convergence to Mirabella’s rising question intonation (Gessinger, 2022). Overall, these findings suggest the possibility that speakers may accommodate to different types of phonetic features depending on their personality structure. However, we encourage future work to look beyond individual phonetic features and explore the predictive potential of the boundary construct with respect to a general disposition to adaptive speech behavior.

Supplemental Material

sj-pdf-1-las-10.1177_00238309251341895 – Supplemental material for Exploring the Relationship Between Mental Boundary Strength and Phonetic Accommodation

Supplemental material, sj-pdf-1-las-10.1177_00238309251341895 for Exploring the Relationship Between Mental Boundary Strength and Phonetic Accommodation by Iona Gessinger, Nicolas Becker and Benjamin R. Cowan in Language and Speech