Silja Ang-Tschachtli - Bilingual Couples in Conversation

), loud ( ), low ( ), high ( ), husky ( ), creaky ( ), smiling ( <:-)>) or nasal ( ) voice. Some changes in pitch and loudness over the course of an intonation unit or turn were indicated as well, namely increasing or decreasing loudness ( and ) and pitch ( and ). In addition, I marked very fast and slow speech ( and respectively). If necessary, comments were added in double brackets and translations in single quotation marks (see “List of transcription conventions” at the beginning of the book for a complete list).

Transcribing was painstaking, due to the length of the interviews (11.25 hours in total), the level of detail of the transcriptions and the fact that there were several speakers whose speech overlappedoverlap fairly often (on average, each participant overlapped 528 times with another speaker).17 Nevertheless, I deemed the prosodicprosodic featuressuprasegmental features detail of the transcriptions necessary, as the prosodic features might contain clues about the speakers’communicative intentions or emotional state. Only prosodic elements that I did not believe to yield any insights for the specific purpose of this study were omitted. These include, for example, secondary stresses or standard pausespauses in between intonation units, which were not indicated unless they were exceptionally long. Such pauses are difficult to compare; as Edwards notes, measuring pauses is only partially meaningful, because “perceived pause length depends not only on physically measurable time, but also on speech rate, location of the pause […] and other factors” (2001: 322). Furthermore, it is often impossible to make a judgement on who is responsible for a pause that occurs between turns (2001: 332).

I encountered a number of difficultieschallengesmethodological during the transcription process, which were partly due to the fact that the couples have very different manners of speaking (e.g. shorter or longer pauses, one or more stresses for intonation units). First of all, it was not always clear when to “break” an intonation unit. An intonation unit has been defined as “a stretch of speech uttered under a single coherent intonation contour” (Du Bois et al. 1993: 47). It is supposed to consist of a nucleus (primary accent) and other syllables, though it has been proposed that an intonation unit is often not restricted to this (Szczepek Reed 2006: 28; Du Bois et al. 1993: 56), which can present problems. Szczepek Reed, for instance, infers from her own transcriptions that “a given intonation unit is by no means restricted to one nucleus, but may carry two or more syllables with equally prominent stress” (2006: 28). Chafe (1994) suggests that intonation units can be delimited by pausing, pitch change, duration or tempo changes, intensity, voice quality or speaker changes. Other scholars, such as Gumperz, have grouped intonation units into minor tone groups, which “delimit a message treated as a component of a larger whole, and major tone groups […] which are more independent” (1982: 110). Yet this distinction is also often difficult to uphold. I considered all of these factors in my own transcriptions, but was still obliged to make an instinctive decision in a number of cases. Despite these challenges, I considered distinguishing intonation units to be very important, as they provide information with regard to turn-taking behaviour (see Gumperz 1982) and are necessary to determine terminal pitch.

Another difficult decision, which is linked to the question of the intonation units, was where to place the nuclear stressstressesnuclear. As Gumperz remarks, “accent placement is for the most part grammatically conditioned” and tends to fall on content words towards the end of the intonation unit (1982: 111). This is not always the case, though, and contrastive accents or deaccentuation may also occur. While a single nuclear stress could be clearly discerned in most intonation units in my interviews, there were nonetheless instances where there were several stresses without a significant difference between them. In these situations, I chose to set several nuclear stresses. In most cases, this correlates either with a rhythmic speech ( ) or marked speechspeechmarked ( ), both of which were also indicated.

Moreover, there were considerable differences with regard to stress as well as vowel lengtheninglengthening between the varieties of English spoken by the participants. Generally speaking, British English seems to be more strongly stress-timed than American English,18 and there is a clearer distinction between short and long vowels. This means that the transcriptions might not be entirely uniform as regards aspects such as intonation, lengthening or pausing, since it is possible that a stress or a pause of similar intensity or length might appear more salient in the context of a conversation in one variety of Englishvarietyof English than in another.

Once transcribed, the interviews were inserted into a software application which assists in the management and analysis of large amounts of qualitative and quantitative data (MAXQDA 12MAXQDA). For the coding and the analysis, I proceeded by topic, as each topic is linked to a different field of previous linguistic work, and required different methods of analysis. For each topic, a separate code system was developed, and all relevant segments were coded in MAXQDA 12. In some cases, these segments took the form of longer extracts which were then analysed quantitatively; in other cases, certain features (e.g. laughter) or individual shorter expressions (e.g. code-switches, expressions of emotions, or swearwords) were coded and assigned to categories. Whenever possible, I tried to follow my data in the creation of a code system, rather than creating categories and attempting to fit my data into them (see also Hay 2000: 727). For instance, I looked at each instance of laughter and tried to determine the main trigger; if none of the categories I had already established was suitable, a new one was created. Some of the coding could be carried out automatically (e.g. backchannel signals and suprasegmental features were auto-coded for each participant), but most of it had to be completed manually. In this case, I tended to use the search function and then to assign segments to the appropriate category.

The main advantages of coding segments in MAXQDA 12 were that I could find and draw together everything belonging to one topic (for the qualitative analysis), and that I was also able to count and compare features easily (for the quantitative analysis). In addition, I was able to revisit each coding, for instance, if I wanted to view the context, or if I decided to subdivide a category. Another major benefit of the system was that I could search, code, and finally analyse features for each speaker individually, as the software is able to recognize which speaker a turn belongs to in a transcript involving several speakers. This was particularly helpful in the analysis of quantitative elements for each individual speaker, such as laughter or suprasegmental features.

Once the coded segments had been retrieved from MAXQDA 12, I began with the analysis. I decided to focus on six main topics that seemed to be of particular interest based on the couples’ conversations and previous research. Each of these topics is discussed extensively in a separate chapter. All the chapters are structured in a similar manner, containing an overview of previous work on the subject, a section of quantitative analysis (in the chapters where this was relevant), a qualitative analysis of the participants’ reports and comments on the subject, and an overview in which I summarize the most important results of the analysis, and compare the qualitative and quantitative parts of the analysis with each other and with previous work. For the chapters containing quantitative analyses, the exact methodology used is described in the respective chapters, since this depended on the subject.