The role of linguistic features in response language choice: the case of new Dutch speakers in Groningen

(1)

The role of linguistic features in response language choice: the case of new Dutch

speakers in Groningen

Gina B. L. Pinas

S3522822

MA thesis

Departments of Applied Linguistics and Frisian Language and Culture Faculty of Arts

Rijksuniversiteit Groningen

Supervisors:

Dr. C. S. Gooskens E. Juarros Daussà, PhD

January 22

^th

, 2020

Word count: 14796

(2)

I declare that this thesis is my own work except where indicated otherwise through proper use of quotes and references.

(3)

Abstract

This thesis set out to investigate to what extent linguistic features influence predicting whether or not a native speaker accommodates to a new speaker of their own language through response language choice (RLC). In the research I made use of a mixed-method approach: measurement of perceptual judgements on new speakers of Dutch’ speech through a digital questionnaire made with Qualtrics software for four linguistic features, and computer-aided calculations of one linguistic feature by making use of the Praat software with the Prosogram extension. Intonation likeliness, foreignness of accent, (non-)colloquial sentence structure, and peculiar word choice were the linguistic features of new speakers’

speech that were measured perceptually. The final linguistic feature was approached by measuring the rhythmic and melodic properties, converting it into an RM-space, and then calculating prosodic distance. The speech sample of the new speakers (L2:NL, n =5) came from participants of a native English language background. The perceptual judgement scores came from native Dutch speakers (n=29). A Mann-Whitney U test found none of the linguistic variables to be significant. The biggest crux appeared to be the small sample size.

Keywords: Response language choice, prosodic distance, intonation, foreign accent

(4)

1. Introduction

It took years to realise that the discrepancy that I felt inside myself was caused, in part, by my identity being denied by those from my close social network. While my receptive linguistic skill in Drèents (a Low Saxon dialect) is sufficient for the receiving role in communication, my productive skill is not. Like many parents, my caregivers also thought that accommodating communicatively to me by conversing in standard Dutch would help me.

Instead, I felt positioned into the ‘outsider’ group.

Even in short-lived interaction with strangers, for me it is a regular occurrence that interactions start off in ‘the wrong language’ in what I consider to be my home country (i.e.

speaking English in the Netherlands). Nonetheless, I am not completely surprised by the relatively common mistake fellow nationals make; physically I do not reflect the stereotypical Dutch appearance (e.g., blond hair, blue eyes). That said, Dutch nationals with an appearance that hints to a Surinamese background are hardly rare in the Netherlands (CBS, 2016).

Surprisingly, preliminary exploration of experiences of foreign friends implied similar outcomes, and seemingly, regardless of appearance. Dutch people believe that their proficiency in English is often sufficient to accommodate to foreigners that appear – for whatever reason – not to have a native-like command of the local language. Even if the foreigners in these situations are proficient enough in the Dutch language for mutually intelligible communication. Consequentially, although the interlocutors would be able to be communicate in Dutch, they often communicate in both their second languages. It was these types of situations and reflexions that got me to wonder what the relationship is between the ability to speak multiple languages and the (unconscious) incentive to adjust for communication.

In today’s globalising and increasingly interconnected world, the experiences of my acquaintances and I are not uncommon. Due to the Schengen Agreement in the European Union, the movement of people has become easier within Europe. Europeans find crossing borders and having to employ English as lingua franca to be normal, regardless of their proficiency (Braunmüller, 2008). Moreover, aside from the European border crossers, there also has been an increase in non-European newcomers (CBS, 2018). Many of these migrants have the incentive to learn the local language, but encounter the obstacles of Europe’s

(7)

linguistic norms regarding English as lingua franca and the fear of communication breakdown while conversing (Grice, 1975). Especially European countries and communities that populate many English users but are not native speakers, e.g. the Netherlands (Booij, 2001; TNS Opinion & Social, 2012), will have its L2 English speakers attempt to

‘accommodate’ to the English language in multilingual encounters.

Myers-Scotton (1983) claims that the choice for code (i.e., language in a multilingual encounter) is rational. According to Myers-Scotton, if a speaker encounters a linguistic code that does not fit the norms of their¹ own speech communities, they make a new decision on which language to respond in. Such a decision is an attempt to negotiate the identity of the interlocutor; ‘Do I want such a person to be part of my community?’ (Sachdev, Giles &

Pauwels, 2013). The practicality of such a systemic model for multilingual encounters would be ideal, if the negotiation that took place could be openly established and known to the interlocutors; ‘I want to be part of your speech community and carry the unspoken cultural obligations that come with it’. However, the anecdotes show it is not that simple.

In reality, it seems that when different languages (and cultural identities) meet, their speakers have a linguistic mechanism that jumps to action. The mechanism prompts the identification of phonemic sounds that allows them to either accept communication or reject communication. Occasionally, communication is simply not possible because the interlocutors do not share a linguistic repertoire. In communities that do have multiple linguistic repertoires, it is often a choice that determines the possibility of communication and the manner of communication (e.g., Eckert & McConnell-Ginet, 1992; Gumperz, 1968;

Holmes & Meyerhoff, 1999). In situations where interlocutors share a linguistic repertoire but are not (yet) in the same community, the language of communication is partly dictated by unconscious processes and input (cf. Fano, 1950).

Few studies concern themselves with the response language choice (RLC) based on linguistic features in multilingual interactions and spontaneous encounters, but the socio- psychological approach is overrepresented (e.g., Fishman, 1965; Genesee & Bourhis, 1988;

Myers-Scotton, 1976). There are different possible approaches; however, I chose to let the

‘linguistics as social science’-argument guide me. In this argument, linguistics is to employ

1 They/them/their/those as deliberate use of a gender-neutral equivalent for he/him/his or she/her.

(8)

scientific methods rather than to “rely heavily on intuition and imagination” (Huang &

Chang, 2008, p. 1819; Sapir, 1929).

My aim is to contribute to the field by focussing on the linguistic input in RLC and investigate whether it is possible to differentiate between the influence of linguistic input and indexical information on RLC in multilingual encounters. In here, the linguistic input refers to all information within the linguistic context (e.g., lexical, phonetical, morphological, phonological) that can be received from the speaker in conversation. Indexical information here is adapted from Peirce’s theory of signs, and refers to all the remaining observations an interlocutor can make about the conversation partner that is assigned meaning (Murphey, 1961). For example, the interpretation of a darker skin colour as ‘foreign to the community’.

To keep the research within the scope of this thesis, I demarcated the study to Groningen, the Netherlands. Hence, the aim of this thesis is to distil linguistic factors that can indicate if a Dutch native speaker will adjust to a new speaker of Dutch in their RLC.

The thesis will be structured in the following manner. In order to give insight into this discussion and improve understanding of more facets that possibly add to the language choice in multilingual encounters, I will define the terms and constructs involved. Then, I move on to the main strands and ideas that currently dominate the discussion, followed by the most relevant earlier empirical work on the subject and to conclude the next chapter I will synthesize befitting hypotheses from the reviewed literature.

Chapter 3 contains the details for the methodology of this study. In there, it is discussed why it was deemed necessary to divide the study into three constituent parts. The first experiment was merely an observation of real-time practices; a tallying of the RLC that new speakers of Dutch were responded to in. The second part focussed on the perceptual judgements of new Dutch speakers that local Dutch have. Sequentially, the third part delivered a measurable result on the factual differences between native and new speakers of Dutch in order to adhere to the ‘linguistics as a science’-argument. The results of the first experiment were juxtaposed with the results of other two parts by using a statistical analysis so that it was possible to calculate the possibility that the RLC could be predicted correctly based on perceptual judgements and factual differences.

The direct results of the experiments are described in Chapter 4. The three constituent parts received their own subsections in this chapter, but were put together to discuss their

(9)

synergy in Chapter 5. In this chapter, limitations to the study are also discussed. I end the thesis with a conclusion in Chapter 6, where you can also find suggestions for further research.

I am optimistic that adding to studies on language identification, code and choice will aid in our understanding of how to better deal with the cultural paradox we have in Europe. It is demanded of expatriates to learn the local language for assimilation purposes (Riggins, 1997;

e.g., Janssens, 2015; Yagmur & Akinci, 2003), but simultaneously we are feeding into segregation by communicatively ‘accommodating’ to English (Extra & Yagmur, 2005;

House, 2003; Oakes, 2001). Leaving us to wonder what factors push people to use English as a lingua franca in multilingual interactions with new speakers of their native language.

(10)

2. Theoretical background

This chapter concerns itself with the literature reviewed relating to the topics relevant to the research on the role of linguistic features in response language choice. As said, I will define the constructs used to avoid ambiguous interpretation (section 2.1). Then I will discuss the main strands and ideas from previous research on identification of language (Section 2.2.1), and the processes in response language choice (section 2.2.2). Including a critical synthesis of what those scholars do not cover. Finally, section 2.3 will conclude how the reviewed literature on linguistic code and choice formulates into the exact hypotheses used for this research.

2.1 Defining important constructs

Before exploring other studies’ theories on language choice and the linguistic factors involved, I will define important constructs for this thesis. The following four constructs are ambiguous to an extent in academia. Therefore, I have (re)defined them for use in this thesis.

2.1.1 Code-switching and switching codes

In a thesis on language choice, it is important to clearly describe what is exactly meant with terms that are often encountered in related studies, such as code-switching and multilingual accommodation, in order to demarcate the ones of relevance. Surprisingly, the academic field of language and communication shows a handful of such sometimes seemingly contradictory terms.

Alternation between two or more languages (be it styles or varieties) by multilinguals is commonly called code-switching. In this definition ‘codes’ are used almost synonymously with language. However, as such it has become an umbrella-term for quite a few different alternation forms. To illustrate, code-switching can occur on a level as small as within a word (Gumperz, 1977; Hoffman, 2014), and as large as between the interlocutors (Hoffman, 2014;

Leppänen, 2007).

Before ‘code-switching’ became an umbrella-term, early research on switching codes strictly viewed languages, styles and varieties to have codes, not be codes:

(11)

“Spectrographic analysis has indicated that the different speech sounds used by anyone speaker have easily distinguishable frequency patterns which are essentially stationary with time. This does not seem to be true for speech sounds used by different speakers. If we consider these frequency patterns as code groups, it appears that different speakers use, in a sense, somewhat different codes. These codes are stored in the brain of the listener who uses in each case the appropriate code. New codes are continually learned whenever new people are met, particularly people belonging to different linguistic groups. This point of view is in agreement with the observation that our ability to understand and the effort required to understand depends on our familiarity with the speaker’s voice. In addition, we are often conscious of ‘switching code’ in our brain, particularly when a change of language takes place” (Fano, 1950; as cited in Alvarez-Cáccamo, 1998, p. 31).

Fano’s early definition fits the current study, because it considers the prosodic linguistic features (e.g., frequency patterns) that distinguish codes within languages, styles and/or varieties, and not just lexical differences.

The experiments of this thesis were based on native speakers’ ability to identify new speakers of the language in first encounters and what that information did to their choice of response language. According to Fano, this would mean that although new speakers use the same language, their use of code might give them away as non-native speakers. Therefore, in this thesis I purposely avoid the use of the confusing code-switching, but instead use switching code, because the focus is on communicatively capable new speakers.

2.1.2 Frequency patterns of code groups

Thus, according to Fano (as cited in Alvarez-Cáccamo, 1998), languages have codes and these codes belong to groups that are discernible by their own frequency patterns. Without getting too technical about the technicalities of sound, a simplified approach to the explanation of frequencies in speech research is that frequencies are the voice pitches (Musiek, 1994). Though not completely true (what we call pitch is the correlating

‘fundamental frequency’ (F0) that is perceived by humans), it is sufficient for understanding that the speed in vibrations of the vocal cords measured in cycles per second is what we perceive as pitch (Nooteboom, 1997). Accordingly, frequency patterns in speech are the

(12)

temporal differences between the sequential pitches. Languages are theorised to have their own particular patterns in how one’s voice is used to pitch syllables (i.e., code).

Usually, these frequencies (F0) are measured in Hertz (Hz), but it is sometimes converted to semitones in speech research. By converting the frequency of pitches to semitones, a comparison can be made to musical tones. Especially in studies of linguistic prosody this seems to be a trend. Although the study by Patel, Iverson and Rosenberg (2006) was used as starting point for the experiments in this thesis, for this study it will not be necessary to use semitones as I will not be making a comparison to musical tones. Therefore, I too measure, analyse and speak of voice pitches in Hz.

2.1.3 Multilingual accommodation and speech communities

In the first paragraph of this chapter, I put accommodation between quotations marks. In language and communication sciences, accommodation implies the “[...] constant movement towards and away from others, by changing one’s communicative behaviour” (Sachdev, Giles, & Pauwels, 2013, p. 394). However, colloquial discourse around multilingual accommodation often sounds as if it is a positive, almost altruistic, phenomenon by definition. Likely because ‘to accommodate’ in everyday use can refer to a willingness to help by adjusting. Although multilingual accommodation might be just that in its intention, it is not necessarily wished for by the other party, leaving it failing to adjust (Sachdev, Giles, &

Pauwels, 2013).

Multilingual accommodation refers to the movement of language in use by either: a) converging, the movement of the use of language to fit the speech of the interlocutor; or b) diverging, moving away from the register of the interlocutor. Therefore, it is important to emphasise that when I speak of accommodation in this study, I refer to Giles’ seminal work and corresponding definition; switching the linguistic code as part of communication behaviour, not the sentiment around it. In practice, this means that whether a native speaker chooses to answer in a lingua franca that is not native to either or in the same language they are addressed, in both cases accommodation is occurring.

Accommodation is the movement of language in use that is usually motivated by the identification of the idiosyncratic language variety. The subtle differences in language are used to perform our identities (e.g., Dovchin, 2011). Therefore, when interlocutors accommodate away from each other, we can assume that they reject the other in their speech

(13)

community, but not necessarily in their community of practice (e.g., Cheshire, Kerswill, Fox

& Torgersen, 2011). After all, members of a speech community share similar linguistic practices; a human aggregate that is characterised by frequent set of linguistic peculiarities that mark the speech variety (Gumperz, 1968). A community of practice is also a human aggregate, however, it is not exclusively characterised by the same linguistic practices.

Created to fit modern times, the community of practice considers the possibility of the formation of communities that share social practices, but not necessarily linguistic practices on such a regular basis that it can be considered systematically recognisable. Therefore, when interlocutors accommodate towards each other, we can hypothesize they share similar enough practice accept each other in their speech communities. Hence, RLC might give some primary insights on the linguistic conditions of the local speech communities.

2.2 Theory and reviewed literature

The following sections will discuss two strands of literature. First, section 2.2.1 concerns studies on what I previously deducted to be a ‘mechanism’ in the introduction. It reviews the processes prior to response. Second, in section 2.2.2, I present the literature on the phase suspected to come when a response language is chosen.

2.2.1 Previous research on the identification of language and (new) speakers

Computational linguistics has seen an increase in interest in native language identification (hereafter, NLI). NLI is a reader’s ability to recognize the ‘native’ language of anonymous writers through the authors’ stylistic choices. Generally, these features follow the patterns of the author’s ‘native’ language (Koppel, Schler, & Zigdon, 2005; Malmasi & Dras, 2018).

A great benefit for NLI research is the option to get quantifiable data. There is no linguistic element that relies on the capriciousness of auditory perception, because it concerns the NLI of written language. Therefore, its strength lies in the extensive list of linguistic features that can be taken into consideration. Early arduous work by Mosteller and Wallace (1963) shows that (function) words can be used to identify the author’s dominant or first language. Some decades later, Cimino, Dell’Orletta, Venturi and Montemagni (2013) identified a set of twenty linguistic features to be helpful in linguistic profiling and NLI.

Other researchers have focussed on different features. These included lexical and structural features such as characters and part-of-speech (e.g., Ionescu, Popescu, & Cahill, 2014).

(14)

Even though it is customary for NLI in computational linguistics to focus on systemic of linguistic features in writing, according to Adeeba and Hussain (2019), NLI is “the task of identifying the first language of a user based on their speech or written text in a second language” (p. 17098). Furthermore, their study suggested its worth on non-English use as well. Adopting their conviction, the linguistic features used for the identification in writing- based studies could correspond with the features used for speech-based NLI. It can be hypothesised that particular word use might help identify the speaker’s dominant or first language, or, at the very least, give away that the speaker is a new speaker of the language.

However, from a practical point of view, this would be very difficult to test with solely behavioural experiments. To illustrate, in conversation, it can hardly be expected from participants to be able to relay whether they (un-)consciously discriminate a ‘native’ speaker from a new speaker based on, e.g., type/token ratio like a computer can.

Luckily, language varieties differ not only in lexical and structures features, but also have distinctive temporal and accentual patterning (i.e., rhythm), and intonation (i.e., melody).

Only relatively recently this knowledge has found its way to NLI (e.g., Hönig, Batliner, Weihammer, & Nöth, 2010; Piat, Fohr, & Illina, 2008; Tepperman & Narayanan, 2008), even though it has been long argued that a way to classify languages is based on rhythm. To illustrate, Germanic languages and English are generally found to be stress-timed, Japanese and Tagalog mora-timed, and Romance languages are generally assumed to be syllable-timed in linguistics (e.g., Pike, 1945; Ramus, Dupoux, & Mehler, 2009).

Without computational methods, there has been a plethora of research into the role of prosody in language recognition. These however, largely seemed to lack the measurable element, and were highly subjective (e.g., Lloyd James, 1940; Pike, 1945).

Nazzi, Bertoncini and Mehler (1998) experimented whether or not infants could recognize their native language from others by measuring the infants’ spontaneous sucking rate on a pacifier. Their experiment revealed that infants could discriminate between language-classes based on rhythm but, not very well within such classes (i.e., stress-timed English and stress- timed Dutch). Though Nazzi et al., (1998) did not mention adults, this does suggest that adults, like infants, perhaps at the very least use rhythmic information for language discrimination.

(15)

While both Dutch and English can be classified into the stress-timed languages they do have prosodic differences. Grabe and Low (2002) developed the normalized Pairwise Variability Index (nPVI) to quantify the average Hz in pitch of neighbouring vowels in a sentence that in turn can be used to determine what sentences rhythmically ‘swing’. Meaning that the nPVI can be used as a tool to measure the variations in unequal durations in rhythm.

Others developed or made use of the standard deviation of consonantal intervals (ΔC) or standard deviation of vocalic intervals (ΔV) per value per sentence (%V) as variables (e.g., Ramus, Nespor, & Mehler, 2009). Even though those studies have used it to compare different languages (e.g., White & Mattys, 2005), or languages and music (e.g., Patel, Iversen, & Rosenberg, 2006), it is plausible that it can also be used to compare ‘native’

speakers to new speakers of the same language. Unfortunately, perceptually speaking, successful discrimination between English and Dutch based on the rhythm variable is argued to be at no more than chance level (Ramus, Dupoux, & Mehler, 2009). In conclusion, there seems to be no added value in singling out rhythmical cues in behavioural and perceptual

studies.

Figure 2.1 Spectrogram of a native speaker uttering the sentence “Weet u waar ik een haarborstel kan kopen?” (Own translation: ‘Do you know where I can buy a hairbrush?’)

Another prosodic feature, the melody of speech, is usually taken to be the pitch properties of sentences (e.g., ‘t Hart, Collier, & Cohen, 1990; Nooteboom, 1997). Safe points of pitch measurements within those sentences is at the widest points in the spectrogram of voice measurements. Those are the locations where most stress is put on the vocal cords. In other words, it is the ‘core’ of that syllable’s sound pitch (see figure 2.1 for an example). Vowels are at the core of syllables, and therefore the interval variability of the pitch of vowels would

(16)

lend an indication of the speech melody (Patel, Iversen, & Rosenberg, 2006). Scholars, especially the ones attempting to compare speech melody with musical melody, have been arguing for an approach that highlight more than just the most salient syllabic pitches as to not ignore the rest of the melodic signals (e.g., Chow & Brown, 2018). The question is whether this is necessary for behavioural and perceptual studies.

Colloquially, we often speak of ‘intonation’ when we refer to speech melody; the trajectory of all pitch variations in communication. Academically speaking this definition is being disputed. Experimental studies opt for bringing together perceptual, physiological (e.g., stress), and mentalistic insights on problems and definitions of intonation, and to inseparably bind paralinguistic functions post-lexically (e.g., emphatic force, happy-sounding) to intonation due to the high correlation between the two that has yet to be fully unpacked (Chow & Brown, 2018; Crystal, 1969; Koch, 2008; Pell, 2000; Swerts & Veldhuis, 2001). In her research on the differences between speech in conversation, interviews and lectures, Bel (2015), like many others, decided to measure intonation by examining mean pitch (mean F0) and pitch range (F0 contour). This approach adheres to the definition that limits intonation to concern fundamental frequencies and is therefore independently measurable. However, in order to examine the extent to which patterns of intonation point to identities (Johnstone, 2015), and therefore the aforementioned affective elements, we cannot rely on disconnecting fundamental frequencies from perceptions.

In case of multilingual communication, interlocutors might perceive awkward ‘intonation’

in speech though its communicative function has come across. Though certain phones of a language just intrinsically have a higher pitch than others (e.g., /i/ in juxtaposition to /a/) (Nooteboom, 1997), it can be difficult for new speakers to actively control their F0 in such a way that it imitates native speakers of that language. Consequently, it is hypothesised that L2 speakers of a language might have difficulty to closely imitate speech melody or in other words ‘crack that language’s linguistic code’.

According to Nooteboom (1997), human pitch perception is remarkably accurate.

Therefore, I deduced searching for a different vowel threshold (i.e., pitch) in L2 Dutch speech’ spectrograms to be useful when attempting to discover whether it affects the NLI and indirectly the initial language choice or switching of codes. Yet, almost a decade later, Patel, Iversen and Rosenberg (2006) found that on average vowel-to-vowel pitch movement can be

(17)

similar between languages (here, British English and French) even though the intervals in one language’s sentences were significantly more variable (i.e. English). Moreover, earlier Nooteboom (1997) already argued that if someone were to:

“[…] imitate the intonation or speech melody […] we obtain a pitch curve that will definitely not be perceptually equal to the original. It will be easy to hear many differences. But yet […] a panel of native listeners can hear whether the imitation is successful in conveying the same melodic impression. Apparently, intonation is organized in terms of melodic patterns that are recognizable to native speakers of the language” (pp. 8-9).

Therefore, to exclusively analyse the melody intervals of native Dutch speakers with new speaker’s Dutch speech would probably not be very fruitful either. But to measure melodic properties perceptually (i.e. intonation) is.

Figure 2.2. Examples of RM spaces. On the left the rhythm and melody of French and English speech and music is displayed. On the right different French and English composers have been plotted in a RM space. The x-axes shows the degree of durational contrast between successive vowels in sentences (measured using nPVI

by Patel & Daniele, 2003), while the y-axes shows the variability of pitch movements between successive vowels in sentences (measured using the coefficient of variation of absolute interval size).

(18)

Nevertheless, Patel, Iversen and Rosenberg (2006) have found a quantitative measure which distinguishes between prosody of two languages: RM space (see fig. 2.2). Putting rhythm and melody of native and non-native speakers on orthogonal axes as an RM-space plot can provide insight into their ‘closeness to nativeness’. Though they have compared speech prosody with instrumental music through RM space, as they suggest themselves, it lends a quantitative method for assessing non-native prosody given that it is determined beforehand how prosodic distance relates to perceptual judgements of the speech.

White and Mattys (2005) have studied the rhythm of ‘non-native’ English spoken by native Dutch and found “that the best predictor of native-accent rating was a measure based on the coefficient of variation of vowel duration in sentences. […] [their method] could be adapted to study the relationship between foreign accent judgements and measures of prosodic distance in RM space” (Patel, Iversen & Rosenberg, 2006, p. 3044).

It is noteworthy how few scholars looked at spontaneous speech. Either participants read the sentences (Ramus, Dupoux, & Mehler, 2009) or it was written text analysis (Mosteller &

Wallace, 1963; Cimino, Dell’Orletta, Venturi, & Montemagni, 2013). White and Mattys (2005) did look into spontaneous speech, but they looked only at the rhythm. For this thesis, I also want to look at the melody to ‘objectively’ measure prosodic distance and intonation to measure the perceptual element (and the other linguistic features that I will discuss in the next section).

2.2.2 Previous research on (response) language choice

I think we can agree that being able to identify speech as non-native does not compel the native to respond in a lingua franca that is not native to them. They have a choice. Thus, I argue that there are likely some steps in between the identification of (non-)native speech and the language in which is responded (RLC). Moreover, we have seen that newcomers can be identified through many different aspects, however, it is still unknown whether they play a different role in the motivation for language choice. Furthermore, as almost every theoretical model on psychological decision-making processes prescribe, a goal is required in the mind of the speaker in order to come to a decision (Jepsen & Dilley, 1974; Oliveira, 2007). But that is assuming that there is a carefully thought out decision process involved…

(19)

Earlier empirical work on variation in linguistic code choice in conversation is relatively limited. In a longitudinal study by Börestam (2015), the response language (i.e., a Scandinavian language, English or participant’s native language) was tallied when Icelandic adolescents were asked for directions in a Scandinavian language. Interpretation of the code- switching strategies led to the belief that the switch from prioritising Danish to English as lingua franca had to do with the emphasis that was placed on the languages in school. At the time of the first survey in 1983, the English language had a less prominent role than it did at the time of the surveys held in 1999 and 2004 (though both languages were taught in school at the time of the first and last survey).This led to an increased majority responding to the interlocutor in English instead of any other intelligible Scandinavian language. Subsequently, her study hints that perhaps the speakers’ attitude towards English changed, more so than just their ability determined the code of choice (Delsing & Lundin, 2005; as cited in Börestam, 2015).

The cause for multilingual accommodation, or the motivation for language choice is based on the premise of similarity attraction (Byrne, 1969; as cited in Sachdev, Giles, & Pauwels, 2013). Simply put, this idea suggests that the more similar one person is to another, the more likely that person is to have a positive attitude or to be attracted to the other. Sachdev et al.

(2013) argue that the same principle goes for linguistic properties, even within the same language. For example, accent and lexis. Hence, it could be valuable to test the perceived linguistic distance based on these properties between the native speakers and new speakers of that language in a study on language choice.

In the same line of reasoning and perspective, Myers-Scotton (1983) proposes “a model which explains variation in linguistic code choice” (p. 115). She theorises that motivations for a switch in language and/or code follows from the ‘markedness’ of that code. Any society supposedly has codes that are marked with certain socio-cultural rights and obligations (i.e.

speech community). And then there are codes that are unmarked; their use does not symbolize the negotiation of those rights and obligations (yet). Thus first the markedness of the code of choice is checked. Second, the interlocutor gets to make decisions about the relationship with the speaker, based on one of four objectives:

(20)

“1. To establish for the first time a [rights-and-obligations] set between speaker and addressee.

2. To affirm a previously negotiated [rights-and-obligations] set as salient for the current exchange.

3. To readjust a previously negotiated [rights-and-obligations] set (and thereby change the hierarchy of currently salient social features).

4. To alter radically a previously negotiated [rights-and-obligations] set (and thereby change membership in the set of currently salient social features)” (p. 117).

This goes for established relationships but also for new encounters. Once the interlocutor makes their decision they use the code of choice to symbolise that decision.

Suppose we take the Icelandic adolescents from Börestam’s (2015) study. Now we know that the language they choose to respond in, the RLC, is likely to be motivated by their attitude towards that language, but also by their ideas on the relationship they wish to establish with the person who asked for directions. Provided that they perceive to be similar enough, they are likely to subconsciously have the goal to establish a relationship of some kind and choose to answer in a code that accommodates the speaker.

The few studies discussed above on language choice have something in common. Most approach the subject from a social-psychological perspective. Thereby focussing heavily on social structure, although it is generally accepted to be an interplay between two concepts;

social structure and human agency. On the one hand we have societal structure that supposedly determines the marked codes of societies or societal attitudes on language, but on the other hand we should have the individual practices that reproduce the very structure that shapes them. Human agency does not refer to isolated action, but the socio-culturally mediated capacity to act. Hence, the interplay between structure and agency. Therefore, if we wish to give agency the credit where it is due, we need to look closely and revaluate the uses of language and linguistic form (Ahearn, 2001). Relevant examples to this chapter; individual judgements on language in practice, decision-making in RLC, objectives for accommodating, inter alia.

These perspectives lend ground to believe that there are different heuristic strategies at work in the phase of ‘choosing’ a response language (Broadbent & Josephson, 1981).

Unfortunately, the linguistic features whereupon they are grounded have been hard to identify

(21)

thus far. Although Johnstone (2015) focussed on how (and not which) linguistic features get linked with codes that stereotype identities and activities, she did briefly mention which features she suspected to be the culprits: “[...] particular words, the way of pronouncing words, grammatical patterns, and patterns of intonation [...]” (p. 632). I chose to continue from here.

2.3 Approach, research question & hypotheses

My main quest for the study was to find out more about the phenomenon that occurs when multilinguals have more than one language in common. The literature review led me to believe that the, though seemingly obvious, linguistic perspective is being snowed under by the social psychological perspective. If we could filter out visible social cues, such as foreignness of appearance, what would be left? Which led me to the following, general research question:

 RQ: To what extent do linguistic features predict whether or not a native speaker accommodates to a new speaker of their own language through response language choice?

Aligning with the aim, as discussed in the introduction of the thesis report, the null- hypothesis has been drawn up. Picking a more specific situation in accordance to the phenomenon of interest sketched at the beginning of this section, helped making the research operational. In Groningen there were plenty of newcomers to be found who learned Dutch at a later age, but generally (still) use English as lingua franca to get around. Besides, I proceeded on the assumption that the compulsory education system abled locals to answer simple questions in English.

As conventional, my null-hypothesis is built on the idea that there is no relationship to be found. If most of my data (results) were to show no relation to the variables, or in other words, if most of the participants will be answered back in a ‘random’ language regardless of the linguistic features procured from the reviewed literature (see chapter 2.2), then this hypothesis will not be rejected. Again, as conventional in social sciences, I put the probability of outcomes on a statistical significance threshold of 5%. The resulting null-hypothesis that follows should help predict whether a Dutch native speaker would adjust to a new speaker of Dutch through language choice based on certain linguistic features.

(22)

 H0: There is no relationship between the linguistic features of new speakers’ speech and the possibility of correctly predicting the switch of code of the native speaker.

As discussed before in chapter 2.2, several linguistic features might influence the (un)conscious processes that determine the interlocutors’ RLC. In the phase of spoken NLI and discrimination, even if it is a new speaker of the language, this could include rhythm &

melody and intonation. In the process of choosing a response language this includes foreignness of accent and linguistic style choice (e.g., idiosyncratic word choice or sentence structure). Therefore it was deemed suitable to use the common tool of measurement regarding foreign linguistic ability, Common European Framework of Reference for Languages (CEFR), double as a tool for comparison between participants of the study. Which led me to the following alternative hypothesis:

 Hα : The linguistic features of new speakers of Dutch levels (A1-C1) have a relationship to the language accommodation of native Dutch speakers.

The following hypotheses help discriminate which of the isolated linguistic features share relationship with the response language choice. The hypotheses are separated based on rhythm & melody (prosodic distance), intonation (perceived prosody), foreignness of accent, sentence structure, and idiosyncratic word choice as derived from the literature review from chapter 2.2.

 H1 : Speaker A (L1:NL) accommodates on average more often towards speaker B (L2:NL) when those L2 intonation is less native-like.

 H2 : Speaker A (L1:NL) accommodates on average more often towards speaker B

(L2:NL) when those rhythm to melody relationship (RM-space) is increasingly further removed from average native Dutch speech.

(L2:NL) when those accent is perceived as foreign.

(L2:NL) when those choice of words is increasingly peculiar.

(23)

 H5 : Speaker A (L1:NL) accommodates on average more often towards speaker B (L2:NL) when those sentence structures are increasingly non-colloquial.

3. Methodology

The following chapter contains the outlines of the approach in which the research was conducted. Section 3.1 discusses method for measuring language response choice in spontaneous speech which, when measured, lend the dependable variable of the research.

Next, in section 3.2, the variables that were hypothesised to have an effect on the response language choice are discussed. This section is dived into two subsection each discussing two different methods. The RM-space method is presented in subsection 3.2.2, and in subsection 3.2.3 a survey. Lastly, in section 3.3, the choice on (statistical) analyses of the data is discussed.

3.1 Language response choice (the dependent variable)

For the dependent variable of this study, language response choice (RLC), a small group of five new speakers of Dutch (L2:NL) were chosen to incite spontaneous speech from local and native speakers of Dutch. The L2:NL were subject to analyses of their speech, and hence, they had to be selected carefully. Ideally, they were of the same language background in order to keep the prosodic elements of their native language that might ‘leak through’ when speaking in Dutch comparable to each other. In such a way, the judgements on the used language would not be subverted to the locals’ judgement of the social status of one’s language background. I managed to gather five L2:NL that were native speakers of English.

Although the same language background was aspired, differences in language proficiency was sought after. There is the credence that those who can mimic the accent and intonation of the relevant language closely are better at that language. Therefore, by spreading the L2:NL’s proficiencies, I inferred it to be possible to determine to what extent proficiency holds a relation to RLC. The L2:NL were asked to report their ‘proficiency levels’ according to the descriptions of the Common European Framework of Reference for Languages (CEFR).

Basic descriptions of each L2:NL-participant can be found below in table 1 on the following page. My sample had a 4 to 1 ratio in favour of women. The participants were ranged 19 to 31 in age, and were aged 24 years on average. All were students of the

(24)

University of Groningen. The participants did not have to meet any requirements aside from their linguistic background and CEFR level. Aside from linguistic backgrounds, possible similarities are purely coincidental.

Table 1.

Demographics of L2:NL-speakers Participan

t

L1 NL CEFR gender age

I ENG B1 F 19

II ENG C1 F 31

III ENG B1 F 21

IV ENG A2 M 26

V ENG A2 F 23

The choice in language response was observed in 24 native speakers of Dutch. Because the aim was to observe language response choice in spontaneous speech interaction, this group was not officially invited to participate. At the start of the task, the L2:NL participants were explained that they were relatively free to choose whomever they wanted but to attempt to approach a variety of people so as to randomise the sample to represent the population.

The small alphabetical letters in the first column of table 2 represent the individuals of this sample of locals (n=24), and the representation of the age groups were well-spread (7 to 9 people per age group). However, women were heavily over-asked in comparison to the men;

double as many women were approached by our new speakers of Dutch (M=8, F=16).

Further descriptions of these locals can be found in table 2 on the following page.

(25)

As briefly mentioned, the L2:NL were tasked to incite spontaneous speech from local and native speakers of Dutch. I prepared for each L2:NL participant to meet me in person separately.

Then, I asked the participant concerned to approach a handful variety of local strangers and ask this person a question in Dutch. I suggested the L2:NL participants, in English, to ask for directions to a location of their choice or where they could buy an item of their choice “or something” for two reasons. First, I hypothesised that by making my suggestion in English I would minimise influencing the L2:NL participants’ Dutch, and thus, keep the interactions to come as spontaneous as possible. Second, I abstained from telling the L2:NL participants exactly what to ask for, for the same reason; to ensure as much spontaneity as possible in their task. Finally, the L2:NL participants did not know about the purpose of the study.

Table 2.

Demographics of the sample of locals

Participants Age group Gender

I II III IV V 15-35 35-55 55+ M F

a. × ×

b. × ×

c. × ×

d. × ×

e. × ×

f. × ×

g. × ×

h. × ×

i. × ×

j. × ×

k. × ×

l. × ×

m. × ×

n. × ×

o. × ×

p. × ×

q. × ×

r. × ×

s. × ×

t. × ×

u. × ×

v. × ×

w. × ×

x. × ×

Total 7 9 8 8 16

(26)

I joined the L2:NL participants as if I were a friend and observed the local interlocutors’

response language choice, guesstimated their age and noted their performed gender after each encounter. Although joining the L2:NL when executing their task risks informing the local that their behaviour was in fact being observed, it was balanced and finally deemed of less risk than obstructing the research by invoking participant bias. Participant biases follow from the need for people to act accordingly; were I to ask for the information needed, chances would have been that I would receive answers participants presumed I wanted. Which, in turn, would be undesirable for the research validity. The task results that yielded the dependent variable, response language choice, can be found in chapter 4.

In order to observe the linguistic behaviour of multilingual people, a study such as this and the one by Börestam (2015) would suffice. However, an experiment with new speakers and native interlocutors engaging in spontaneous speech on the streets as described, gives a limited insight into the linguistic features that are involved in response language choice. After all, the locals’ motivation to switch code could be affected by assumptions based on non- linguistic features (e.g., outward appearance, attitude, goal) (Börestam, 2015; Myers-Scotton, 1983; Sachdev, Giles, & Pauwels, 2013). The following section discusses the linguistic variables that were taken into consideration for this study to move beyond the observation of linguistic behaviour.

3.2 Oral linguistic features of new speakers (the independent variables)

It was hypothesised that perhaps oral linguistic features of (new) speakers play a significant role in language response choice. From the reviewed literature, I identified and demarcated the following linguistic variables: the prosodic distance between the varieties of the interlocutors; atypical intonation; foreignness of accent of the new speaker;

(non-)colloquial sentence structure by the new speaker; and idiosyncratic word choice. In order to determine the significance of the correlation between these linguistic features and language response choice in spontaneous speech, we recorded the five speakers. This section discusses them as the independent variables of the research.

(27)

3.2.1 Representative data of the new speakers

Recording the L2:NL participants in order to analyse their semi-spontaneous speech did not fit the task that yielded the dependent variable for this research. Visibly recording during the interactions could have obstructed the spontaneity of the speech and therefore create a research bias, and secret recording was deemed unethical. Hence, the L2:NL were asked to be recorded in a studio.

The same five L2:NL speakers were recorded speaking. Recording was done on a microphone in mono at 44100 Hz in a sound-dampening recording room at the University of Groningen. To get data as closely as possible to spontaneous speech, every L2:NL was asked in English to utter one sentence in Dutch as they would do on the streets asking a stranger for directions or where to buy an item. Their recorded speech samples assumed as representative data to test with the independent variables of the research. It can be found transcribed in writing and in prosogram in Appendix I.

Table 3 tells the basic descriptive statistics of this data: e.g., the mean duration per sentence measured in seconds; the speech rate calculated in syllable per second, and the mean average fundamental frequency of the sentences.

Table 3.

Descriptive statistics on the 5 different sentences studied.

L2:NL Dutch sentences (n=5) Duration (s) mean (sd) 2.12 (0.33)

Speech rate (syll/s) mean (sd) 4.88 (0.64) No. vowels / sentence mean (sd) 9.4 (1.34) Avg F0 (Hz) mean (sd) 221.2 (52.29)

Total no. vowels 47

Note. Data presented as: average result (standard deviation) 3.2.2 Prosodic distance

From the literature review, five linguistic features were chosen to test their correlation to whether or not a native speaker would ‘accommodate’ to a new speaker of their own language. Based on previous research on the identification of language and (non-)native

(28)

speakers (chapter 2.2.1), it could be deducted that isolating melody or rhythm as separate features of non-native speech would not be fruitful because the prosody of Dutch and the L2:NL’s linguistic background language, English, are too akin to each other. However, if both would be plotted together in Patel, Iversen and Rosenberg’s (2006) Rhythm-Melody space (hereafter, RM-space), it could show the prosodic distance of that non-native speaker in juxtaposition to a native speaker. Hence, rhythm and melody, or rather the prosodic distance when compared to another language variety or register, was a linguistic feature taken to be an independent variable.

Currently, there are not many alternatives to measure non-native prosody. Moreover, as discussed in chapter 2.2, to quantify non-native prosody it appears to be superfluous to compare just the rhythm of Dutch and English, because successful discrimination between the two in natural speech is at no more than chance level (Ramus, Dupoux, & Mehler, 2009).

Furthermore, to measure solely melody, would not say enough about the experience of intonation. The concept of intonation that is commonly known, and can be asked about through a questionnaire, has a different semantic referent than solely melody (Chow &

Brown, 2018). Intonation refers to a wider set of prosodic features of which melody is one (see literature review 2.2.1). Hence, again, it would have been difficult to determine the result’s validity based on just melody. Despite their lack of purpose in isolated form for this research, when the properties are plotted together we see the characteristics of non-native speech. In Patel, Iversen and Rosenberg’s (2006) study that yielded significant results between closely related languages. Thus, the prosodic features of this research that are difficult to be perceptually judged by listeners were ‘objectively’ measured using an RM- space as well.

The rhythm of the speech was quantified by using the normalized Pairwise Variability Index (nPVI). When comparing languages that are differently rhythmically timed (e.g., stress-timed vs. syllable-timed), the outcome would provide us with an indication of how natural the utterances feel based on the temporal patterning of the vowels (Patel & Daniele, 2003). However, a said before, between English and Dutch specifically, these findings were not expected to surpass chance-level. Nevertheless, the nPVI was calculated to be used for the RM-space using the following formula:

(29)

nPVI =100 n−1×

∑

k=1

n−1 d_k−d_k+1 (d_k+d_k+1)/2

In this formula, n is the number of syllables, and the d indicates the duration of the k^th interval (dk) (Deterding, 1994; Low, Grabe & Nolan, 2000; Patel & Daniele, 2003, p. B37).

The normalized PVI accounts for the effects of speaking rate. As this study was coined with the initiative to simulate spontaneous speech situations, it was important to choose the nPVI (cf. White & Mattys, 2007). Although the difference is expected to be small when one omits the arbitrariness that comes with setting the boundaries between vocalic intervals (Deterding, 1994). For the sake of following Patel, Iversen, and Rosenberg’s (2006) RM-space, the successive vocalic intervals were chosen as values in the formula (nPVI-V), not the syllables, as could be argued to be more suitable for measurements of natural conversational speech (cf.

Deterding, 1994). That way it would allow me to stay true to how the RM-space was suggested to be used according to Patel, Iversen, and Rosenberg (2006).

The RM-space was made by also measuring melody in melodic interval variability (MIV).

MIV is a 100 times the coefficient of variation (CV) of pitch interval variability, so that the scaling fits the value range that the nPVI is usually measured in. However, before that, it would be necessary to calculate the pitch interval variability of the sentences from the corpora by dividing the standard deviation (sd) by their mean (M) (Patel, Iversen &

Rosenberg, 2006). These values, in turn, were taken by transcribing the pitch in a prosogram in Praat (Mertens, 2004). A prosogram is in principle similar to the stylization of raw fundamental frequencies but it takes pitch as it is perceived by humans into account (refer to fig. 4.2 below for an example).

(30)

Figure 4.2 is adapted from Patel, Iversen and Rosenberg (2006). It shows the practical difference between successive voice pitch stylized using fundamental frequencies versus Merten’s (2004) prosogram.

Thereafter, the prosodic distance (pd) between native Dutch speech and the L2:NL utterances was calculated. Representative data on the prosodic patterns of native Dutch language speakers (L1:NL) comes from a study by White and Mattys (2005; 2007). Their findings put native Dutch speech at nPVI = 82, and MIV = 65. In the following formula L2 stands for the audio recordings of L2:NL participants and L1 for native Dutch speech.

pd ( L1 , L2 )=

√ ⁽

^nPVI^L1⁻^nPVI^{L 2}

⁾

²⁺

⁽

^MIV^L1^−MIV^L2

⁾

²

At the end of this experiment, we would have quantified data of L2:NL prosody, one of the five linguistic variables that were hypothesised to have part in a native Dutch speaker’s linguistic accommodation behaviour (i.e., response language choice: Dutch/English). The results of this constituent part can be found in chapter 4.2.

3.2.3 Perceptual judgements of L2:NL speech

In order to measure the correlation between L2:NL participants’ linguistic features and response language choice a sample of native Dutch speakers (hereafter, the L1:NL group) were gathered. L1:NL were given the audio recordings of the five L2:NL participants. The local native Dutch speaking interlocutors on the streets were not bothered to fill in a questionnaire after their interaction with the L2:NL participant.

The L1:NL participants were only required to have Dutch as their dominant mother tongue. Ideally, they would have been a representative sample of the population (e.g., spread in age according to the make-up of the Groningen city population; nearly 50% of the female sex, etc.). However, nearly 30% of the initial participants were not fit to be considered. They were not local to Groningen but filled in the questionnaire anyway due to the accessibility of its digital nature. Table 4 shows the demographics of the L1:NL participants providing their perception of the L2:NL speech samples (n=29, M=7, F=22; Modeage=15-34).

Table 4.

Demographics of audio-listeners

Participant ENG CEFR gender age

3 C2 F 15-34

4 C2 M 15-34

5 B1 F 15-34

7 B1 F > 55

(31)

8 C2 M 15-34 Table 4. Demographics of audio-listeners continued.

9 C2 F 15-34

10 C2 F 15-34

11 C1 M 15-34

12 C1 F 15-34

13 C1 F 15-34

14 × M 35-54

15 B1 F 35-54

16 B2 M > 55

17 C1 F 15-34

18 B1 F > 55

22 C2 F 15-34

23 B2 F 15-34

24 B2 M 15-34

25 C2 F 15-34

26 C1 F 15-34

27 B1 F 15-34

28 C1 F 15-34

29 C2 F 15-34

30 B2 F 15-34

31 B2 F 15-34

32 B2 F 15-34

33 B1 F 15-34

34 C2 F 15-34

35 B2 M 15-34

There seems to be a gap in measurable knowledge on how native speakers recognize fellow native speakers based on only linguistic features. Hitherto, it seemed to be a largely intuitive matter. Thus, the L1:NL participants were given the task to rate their perceived feelings regarding the L2:NL speech in order to gather measurable data of their judgements on four linguistic independent variables: (a) awkward intonation, (b) foreignness of accent of the new speaker, (c) (non-)colloquial sentence structure use by the new speaker, and (d) idiosyncratic word choice. By listening to an audio-recording of L2:NL speech, the L1:NL participants were blinded to any visual input that might have had an influence on their perception of the speech samples. In this way, it was attempted to bypass observer biases.

On these recordings were the clips in which the L2:NL participants asked a question as they would ask it to a stranger on the street. After listening to the recordings, the L1:NL filled in a questionnaire about these recorded questions. Questions with 7 point-Likert scales were

(32)

used to give an emotional rating. There were a total of 17 questions of which 8 were on their personal perceptions of the recorded speech (limited to the four linguistic variables), and 5 were on their imagined linguistic behaviour when faced with the question on the street. The questionnaire was created and distributed digitally using the Qualtrics software. The content of the questionnaire can be found in Appendix II.

Every linguistic variable was directly linked to two questions on the questionnaire, and the last four questions served as supporting information. The questionnaire is straightforward and was piloted twice. For their perceptions on the linguistic features in L2:NL speech I asked the L1:NL participants to rate their personal attitude and their perception of the societal attitude per linguistic variable. Question 1, 4, 8, 9, 10, and 11 were asked reversely. In other words, a high score on these questions testifies of a negative view on the L2:NL’s ‘nativeness’. The five questions (Q9-Q13) on their self-reported linguistic behaviour were inserted to compare with outcomes of the research. By asking the L1:NL participants to imagine their linguistic behaviour when faced with the question on the street, we would receive an indication whether their perceptual judgements reflect the results of the RLC. The four questions on their demographics (Q14-Q17) were inserted to serve as supportive information in the discussion of the research outcome. These questions concerned participant age, gender, English CEFR- level, and native linguistic repertoire.

It is of importance to realise that a speech sample with a mean score of 7 on a tested linguistic feature would not necessarily mean that the L2:NL participants’ speech is ‘perfect’.

It means that the L1:NL speakers perceived the speech sample to be very normal and speaks of acceptance to their (hypothetical) speech community.

3.3 (Statistical) analysis of the research

Finally, in this section I briefly summarize the research design and the process of analysis.

Figure 3.1 shows the core of the research design. The dependent variable is depicted on the left side of the figure. The four independent linguistic variables that were perceptually measured through surveying are on the right in the box-shapes, and the independent linguistic variable that was measured through ‘objective’ analysis is displayed on the right side of the figure in an oval shape.

(33)

Figure 3.1 is a conceptual model of the linguistic variables possibly affecting the language response choices.

A multivariate logistic regression analysis would have aided in estimating the possible associations of the independent linguistic features on the binary outcome; response language choice. However, the data did not meet the assumptions to reach a valid outcome for this type of analysis (i.e., insufficient sample size, non-independency of samples). Moreover, a dataset acquired according to the methodology as described above asks for integrating a complex clustering of the sample within the statistical analysis. The data units acquired from measuring the independent variables would need to be grouped because the same factors were repeatedly checked per participant (i.e., 5⁵experimental conditions). Additionally, it would need to be taken into account that it is likely that the linguistic factors are somewhat interactional. Lastly, it is established that in order to make reliable estimations of associations for the individual independent variables in a regression model, at least ten cases with the least frequent outcome of the binary outcome are required per independent variable (Peduzzi, Concato, Holford, & Feinstein, 1996). Therefore, the acquired sample size was too small to be able to reliably perform a multivariate regression analysis and therefore does not warrant this approach.

Hence, I chose to do a univariate analysis. This statistical analysis requires less strict assumptions, and it involves only one variable per test. In this manner I was still able to explore the relation between the individual linguistic variables on the response language choice. While this method does not assess the independent variables simultaneously, thereby limiting analysis of interaction patterns between the variables or controlling for confounding

Response language choice

prosodic distance intonation foreignness of accent

(non-)colloquial sentence structures particular word choice

(34)

descriptive statistics on the ordinal data, it can be qualitatively interpreted and inferential statistics may provide an insight into variable significance.

4. Results

This chapter concerns itself with the results of the linguistic behaviour observations, prosodic analysis and surveys, of which the methodology was explained in detail in the previous chapter. First, in section 4.1, you will find the results of the observations on the streets that yielded the data for the dependent variable. Second, the ‘objective’ data of the L2:NL speech samples on rhythm and melody; analysis using the Praat software will be presented in section 4.2, followed by the results of prosodic distance calculations. Third, in section 4.2.2, you will find the results of the questionnaire survey. Finally, the statistical results of the complete research can be found at the end of this chapter in the last section: 4.3.

4.1. Accommodation to L2:NL in response language choice

Table 4 on the following page presents the language response choice of the observations on the streets. Again, the first five columns (I to V) represent the same five L2:NL participants that had their speech performance in Dutch analysed. The letter a to x in the rows under these columns represent the – by the L2:NL participants chosen – locals on the streets of Groningen that were subject to observation in their interaction with the L2:NL participant I to V. In the two columns on the right you will find the language they decided to answer L2:NL participant I to V in though they were addressed in Dutch.

The results of table 5 yield the values for the dependent variable in the binary prediction.

Regardless of their CEFR-level, the L2:NL spontaneous speech received a response back in Dutch for the majority of the time (83.3%). Participants III and IV were not once answered in English. Of the 24 responses, only four responses were in English. All four responses in English for the L2:NL were given by women above 35 years old (three 55+, one 35-55).

(35)

4.2 Results measurement of linguistic variables

The results of the measuring of all five independent variables, i.e., the linguistic features investigated, are presented in this section. The sections are subdivided based on method of attainment of results.

4.2.1 Perceptual judgements on L2:NL speech

This section presents the results of the survey on the perception of the five L2:NL speakers. The seven options on the Likert-scales of the questionnaire were converted into a median score per L2:NL participant per linguistic feature and can be found in table 6. The judgements come from a sample of the population containing 29 L1:NL participants.

Preliminary experiment: dependent variable

Participants Response

I II III IV V NL (1) ENG (0)

a. ×

b. ×

c. ×

d. ×

e. ×

f. ×

g. ×

h. ×

i. ×

j. ×

k. ×

l. ×

m. ×

n. ×

o. ×

p. ×

q. ×

r. ×

s. ×

t. ×

u. ×

v. ×

w. ×

x. ×

The role of linguistic features in response language choice: the case of new Dutch speakers in Groningen