James M Scobbie and Jane Stuart-Smith
2012
Socially stratified sampling
in laboratory-based
phonological experimentation
The Oxford Handbook of Laboratory Phonology
A.C. Cohn, C. Fougeron, and M.K. Huffman (eds.)
Oxford University Press, Oxford
pages 607-621, chapter 21.1
ISBN 9780199575039
This is an open access, unformatted, author version, almost
identical to the published version.
Please consult the published version for correct pagination,
formatting, and references
http://eresearch.qmu.ac.uk/2281/
The first pages of the chapter, giving context and titles of sections 21.2 and 21.3
Ch 21 Experimental Design and Data Collection
21.1 Socially stratified sampling in laboratory-based phonological
experimentation
21.1.1 Introduction
There has been a division in phonology for several decades between sociolinguistic
research, with a theoretical focus on variation and change at the level of the community,
and generative research, which has taken a purportedly cognitive perspective. These
different theoretical interests are not logically incompatible, of course, and hope springs
eternal for positive interaction, since individual speakers encode aspects of variation as
part of their own phonological competence, and open-ended and overlapping sociallyembedded patterns emerge from a collection of individual grammars. Since, at the most
fundamental level, it is difficult to disagree with Labov’s observation that there could
never be “a successful linguistic theory or practice which is not social” (Labov, 1972:
xiii); the challenge is rather how to go about constructing such a theory, and on what
grounds.
The different theoretical perspectives come complete with their own
methodological practices and insights, and professional specialization by linguists tends
to hamper cross-over and the advancement of understanding in areas of joint interest.
Even sociolinguists and laboratory phonologists display a reticence to adopt any aspects
of methodology from the other camp by those interested in what can be called core
theoretical questions (rather than interface issues) due to a misplaced belief that theory
and methodology are mutually indissoluble. Yet at heart, sociolinguistics and laboratory
phonology are deeply similar empirical quantitative disciplines: Labov’s early work in
the 1960s in Martha’s Vineyard should be seen by laboratory phonologists as part of the
history of their own sub-discipline in terms of methodology. After all, it was chosen “as
a laboratory for an initial investigation of social patterns in linguistic change” (Labov
1972: 4) in order “to avoid the inevitable obscurity of texts, the self-consciousness of
formal elicitations, and the self-deception of introspection” (Labov 1972: xix).
Explicit experimentation (i.e. manipulating speaker behaviour in a premeditated
task-based manner to elicit particular measurable outcomes) is also a fundamental
method in sociolinguistics, as indicated by Labov’s canonical “fourth floor” piece of
experimental fieldwork (Labov 1972). Despite this, laboratory-based or highly technical
experimental studies in sociolinguistics are less usual than the more central variationist
methods of interview and spontaneous speech analysis (but see e.g., Thomas 2002;
Campbell-Kibler 2007; see also Warren and Hay, this chapter, Docherty and MendozaDenton, this volume). Of course, analysis (e.g. transcription or acoustic analysis) of a
“sociolinguistic wordlist” is actually a simple experiment.
This section explores the adoption of one main aspect of sociolinguistic
methodology to see what it can offer to the traditional experimental or theoretical
phonologist for whom formal elicitation of key low-frequency forms remains the order
of the day (though the same lessons apply to, say, analysis of spontaneous corpora (see
Warner, this chapter). Specifically, we propose the use of social or dialectal
stratification of participants as an integral part of laboratory-based research. After
reviewing some basic issues and exemplifying some of them from our own work, we
will provide an initial checklist of methodological ideas.
21.1.2 The fallacy of the neutral participant
It is well-known that subjects or participants in social science experiments often have to
be blinded to the goals and methods of the experimenter, because knowledge of the
purpose of a particular experimental task often alters the behaviour of the participant.
The “Observer’s Paradox” (Labov 1972) has been an important topic for debate and
discussion in the context of sociolinguistic methodology, the aim of which has been
characterised as studying “how people talk when they are not being systematically
observed [despite the fact that] …we can only obtain these data by systematic
observation” (Labov 1972: 209). In sociolinguistics there is a strong interest in the most
vernacular, least standard, most unmonitored forms that can be observed, since it is
argued this is where variation is strongest, thus providing the most evidence for
theoretical investigation (for discussion, see Milroy 1987; Bucholtz 2003; Coupland
2007: 180ff). (By definition, non-vernacular varieties comprise standardised and less
variable systems, which therefore provide less data for variation-oriented theoretical
investigation.) In the psycholinguistic or phonetic laboratory, avoidance of such a
paradox may be part and parcel of experimental design, but it is just one concern
amongst a number of possible problems.
Typically, in the laboratory, unwanted participant bias is avoided through
recruitment processes and briefings which mask, misdirect from, or do not mention the
real purpose of the experiment; and through the use of protocols, tasks and materials
which often distract and conceal from inherent patterns in the design. A more recent and
more sophisticated approach is to manipulate not merely experimental tasks and
materials, but to treat participant knowledge, both explicit and implicit, as a
conditioning factor, in order to see what effect different levels of knowledge have on the
task (see Warren and Hay, this chapter). Contrasting implicit linguistic knowledge and
behaviour against explicit knowledge in turn enables the study of, for example, the
salience of sociolinguistic variables, or the relationship between different levels of
meta-linguistic awareness (e.g. Campbell-Kibler 2007).
Generally speaking, laboratory-based approaches to phonology (Ohala and
Jaeger, 1995; Pierrehumbert, Beckman and Ladd, 2000/this volume) downplay and even
reject traditional non-empirical methods in theorization. So there is no more
introspection about contrast, alternation and identity by the lone phonologist and little
emphasis on fieldwork in collaboration with a small group of native-speakers. Instead,
experimental and quantitative research occupies the methodological centre ground,
enjoying particular success when these methods have been used to probe areas of
grammar in which it is most clear that introspection and even self-recording are
unreliable. The study of “low-level” effects provides data crucial to our understanding
of pretty much everything in phonological systems more subtle than bare phonemic
contrast. Understanding such phenomena requires the study of relatively impromptu,
unmonitored, natural speech which illuminates crucial (and often rarely used) word
combinations and phone sequences, often with prosodic variants of the same. Yet on the
whole these laboratory experiments rely on participants who are highly-intelligent,
fluent, literate, and who are often colleagues or students (see Docherty and Foulkes
2000). Easy recruitment of research subjects is one reason for this bias, particularly
when the research is tedious or uncomfortable. In the case of English, convenience
sampling tends to involve speakers of Standard English: and though English may well
be the most experimented-on language to date, its standardised forms are merely one
narrow aspect. Indeed, restricting research to standard varieties lays the field open to the
criticism that the effects of literacy, education and prescriptive attitudes are filtering
through into the data unobserved or unquantified.
So, though things are changing fast, a substantial proportion of earlier work in
laboratory phonology is open to one of the criticisms that was levelled at generative
phonology: namely that it does not exploit socially and dialectally mediated variation as
a methodological tool for narrow phonological theoretical concerns. And, of course, it
cannot have anything significant to contribute to a wider and more inclusive theoretical
understanding of the individual’s representation and implementation of a grammar
which encodes (whether modularly or holistically) their “own” narrow generative
system as part of the social system around them (Foulkes and Docherty, 2006; Docherty
2007). We need to reject the idea that we can study, impressionistically or
experimentally, some idealised neutral speaker of a variety without understanding better
that “a linguistic variety” always includes variation.
21.1.3 Structured heterogeneity in the sampling of speakers and listeners: from crosslinguistic to cross-dialectal designs
There is an assumption in generative research that any single speaker of a language is a
complete topic for study, yet for reasons that are unclear, there is often no interest in
exploring differences between the numerous minimally-different systems that exist,
despite the opportunities this provides for a more subtle version of cross-linguistic
research (Scobbie, 2007). Given that participants are not neutral and the population to
be sampled is not homogeneous, the standard “homogeneous” sample will always
exhibit interspeaker variation. This can be ignored as noise through pooling and
statistical analysis or attributed to a combination of noise, investigatable artefacts (like
speech rate) and indexical setting (as in the VOT study of Allen, Miller and DeStono,
2003).
A quite different approach is to recruit experimental subjects who are known to
vary, perhaps along such traditional sociolinguistic lines as sex, age, social class or
geographical micro-dialect, or to use structures and practices identified through
ethnography (e.g. Drager 2009 and Warren and Hay, this chapter), or some combination
or extension of these (e.g. the rather different approach to VOT of Scobbie, 2006).
Whether the sample is intentionally bimodal, multimodal or even continuously varying,
the structure of the participant pool can begin to explain some of the noise. Such
participant stratification is hardly an innovation: speaker sex is a common factor used in
experiments due to the actual or potential physiological effects of speaker sex on speech
production (or age, hearing loss, or medication, and geographical dialect differences of
a rather coarse grain are sometimes exploited — but what is extraordinary is that social
differences are exploited only very rarely, despite the opportunities they provide. It is
therefore of particular theoretical and methodological interest when speakers exploit
physiologically-related differences, e.g. sex, for social ends (Stuart-Smith 2007).
Social stratification extends the traditional cross-linguistic design (e.g. the VOT
study of Cho and Ladefoged 1999) to what are essentially different accents of the same
language (which we can call cross-dialectal for convenience). For perception studies,
see Warren and Hay and Warren (this chapter). For cross-dialect research there is no
need to randomly sample on non-linguistic grounds, as a sociolinguistic study might,
where sampling the proportion of different behaviour in the community is important.
(For different possible methods of sampling in sociolinguistics, see e.g. Milroy and
Gordon 2003, and see Trochim and Donnelly 2006 for a more general review). But for
the laboratory phonologist, it may be enough that certain patterns exist, though knowing
something about the prevalence of different patterns is indeed preferable. It seems
obvious that an investigation of French would not randomly sample Western Europeans,
or even people living in France: French speakers would be recruited! It somehow seems
more contentious to take this approach in the study of highly vernacular sociallydefined varieties of a language. Yet of course a laboratory-based phonological study of,
say, spirantisation of stops must include, by design, speakers who exhibit that
phenomenon. They can then be compared to more standard speakers who are likely to
not show it at all. In such diversive sampling methods it is necessary to define each
speaker group and to accept that there will be random variation within the groups, but
representativeness is an extra condition which is optional, not obligatory. Such groups
each have their own accent or dialect, whether they vary along physiological parameters
(e.g. age or sex), geographical differences (traditionally the remit of dialectologists), or
social ones. A combination of approaches can be taken. Diversity sampling provides a
number of specific speakers of different linguistic varieties, and snowball sampling uses
them to help recruit new experimental participants from their own networks (e.g. Milroy
and Gordon 2003: 32).
Cross-dialectal research avoids an obvious confound which affects crosslinguistic research, which is that different languages by definition have different
lexicons, and probably different phonologies, prosodic, morphological and syntactic
systems. Thus looking at purely phonetic/phonological effects across languages (a
central goal of laboratory phonology) is actually very difficult, because experimentally,
too many factors are changing at once. Research into related accents (which we can for
convenience call “cross-dialectal”), on the other hand, is ideal for laboratory phonology
because dialects tend to vary in sound system, to greater and lesser extent. Ideally, the
same materials can be used. Such variation frequently crosses boundaries in phonetic
space that opens up a more complex understanding of contrast and categorisation
(Scobbie, 2006). One major limitation is that the extent of variation is confined to
genetically related systems, so does not reflect the range of cross-linguistic differences
that exist.
Crossover research faces the challenge of satisfying the methodological
requirements of more than one field, but when it does successfully it can explore
fundamental theoretical issues in more than one domain simultaneously, e.g. in both
sociolinguistics and phonetics (e.g. Foulkes and Docherty 2006). Our argument is that
traditional laboratory-based and indeed consultant-based phonetic and phonological
research can also benefit from social stratification and cross-dialectal designs, even
within the narrower remit of research without specific or explicit sociolinguistic
theoretical goals.
Recruiting participants by making explicit reference to their social dialect (e.g.
broad Glaswegian) rather than their language (e.g. English) naturally exacerbates the
Observer’s Paradox. However, if the experimental task is kept homogeneous across all
participants, then comparisons are still possible if variation is induced by knowledge of
direct observation. For many laboratory phonologists, it will not be a problem that some
participants are less vernacular than they might be in unmonitored spontaneous speech,
or even if they exaggerate vernacular features in a type of hyperspeech. However,
further fundamental methodological research in this area is required, not least because it
is actually of great theoretical interest, since the effects of social dialect cut across
differences in task and style in production (Wassink, Wright and Franklin, 2007) and in
perception (but see Warren and Hay, this chapter), both of which are fundamental to
phonological theory. Laboratory-based studies into socially-varying phonological
categorisation of phonetic substance or core phonological phenomena such as
phonotactics, alternation and contrast are an area ripe for future study.
21.1.4 Case study A. Part 1: Social vs. contrastive functions of phonetic correlates of
variants of Scottish English /ai/
In this section we exemplify some of the points above, drawing on our own research
into Glasgow English (e.g. Stuart-Smith 1999, 2003, Stuart-Smith, Timmins, and
Tweedie, 2007). Thirty two Glaswegian participants were stratified into cells of four,
the pool halved by male/female, older/younger and working class/middle class
parameters. We included a mini-experiment on the Scottish Vowel Length Rule
“SVLR” (McKenna 1988; Scobbie, Hewlett and Turk, 1999; Watt and Ingram 2000) to
probe possible phonemic splits, by supplementing a standard sociolinguistic wordlist
with appropriate materials (see Stuart-Smith 1999 for details).
The SVLR basically describes short and long allophones, but includes an
environment with a “quasi-phonemic contrast” (Scobbie and Stuart-Smith, 2008),
namely before tautomorphemic (word-final) /d/ vs. suffix /#d/. Our socially stratified
sample, even with only a few tokens per speaker, confirmed the finding of the more
traditional experimental study of McKenna (1988), which had more tokens and contexts
but whose participants were only a few university students (Scobbie, Hewlett and Turk
1999). Only three vowels are affected: the monophthongs /i/ and /ʉ/ (need vs. kneed and
brood vs. brewed) and the diphthong /ai/ (side vs. sighed). The monophthongs were
longer in duration by 50% or more before /#d/ relative to /d/.
The diphthong /ai/ was more complex phonetically. The allophonic and quasiphonemic variants of /ai/ (short [ʌ̟ ı] or [əı] and long [ɑ̽ ːe]) differed in three ways
(Scobbie, Turk and Hewlett, 1999). These were (a) duration, (b) the location of the first
(but not second) mora in F1/F2 space, and (c) the relative timing of the transition.
Typically, long vowels are about 25 percent greater in duration; have a first mora that is
lower and backer; and have a transition between the two moras that is later. To see the
effect of the latter two correlates in our experiment, consider Figure 1 (i.e. Figure
21.1.1), showing pooled data (based on the grand means of each sex-age cell’s mean for
side vs. sighed and tide vs. tied) over the first 200ms of the vowel (cutting off some of
the end of longest /ai/). The first half of the long variant of /ai/ is low and back followed
by transitional raising and fronting. Only about a quarter of the short variant precedes
the transition, and up to half its duration occurs near the second, offglide target.
F2 (Hz)
2750
2250
1750
1250
250
350
450
650
F1 (Hz)
550
750
short
long
850
950
Figure 1 Quasi-phonemic variants of /ai/ before /d/ (short) and /#d/ (long) in 32 Glaswegian speakers,
showing a low back first mora and an offglide trajectory of raising and fronting. The circle tickmarks
indicate duration in 50ms intervals over the first 200ms of the vowel (approximately quartiles).
We could investigate these possible cues to the distinction through traditional
laboratory-based measurement, from either a perception or production point of view,
using statistical analysis (e.g. Gordeeva, 2008). Instead, our stratified subject pool
reveals that one possible cue, the quality of the first mora, varies socially (Scobbie, Turk
and Hewlett, 1999 and further unpublished work). In side vs. sighed or tide vs. tied, all
speakers distinguish SVLR length using all the correlates mentioned above, but in
addition, MC speakers in general seem to have lower/backer targets (for both short and
long variants) than WC speakers (Figure 2). The first mora quality in F1/F2 space is
therefore unlikely to be an important cue to the short/long length distinction.
The quality difference in the first mora between long and short variants, is, we
think, due to target undershoot in short /ai/ caused by its reduced duration. Given time
constraints, the articulatory system tends not to lower fully, failing to reach the target. A
lesser duration for the first mora could have knock-on effects for overall duration and
the timing of the transition. Lengthening/shortening affects the start of the vowel.
Figure 2 shows the short vs. long SVLR difference and the class difference in real time
over the first 200ms of /ai/, pooling gender and age.
F2 (Hz)
2600
2400
2200
2000
1800
1600
1400
250
350
550
mc short
mc long
wc short
wc long
F1 (Hz)
450
650
750
850
Figure 2 Quasi-phonemic variants of /ai/ with marks at 25ms intervals, showing a social difference in the
F1/F2 location of the first mora of the short variant.
We think that the SVLR short/long difference in /ai/ is less likely to be cued by
raw duration: more important is the relative timing of (the start of) the diphthong
transition form the first mora to the second: all eight speaker groups have a relatively
early transition for the short variant and a relatively late one for the long variant. Also,
though normally the transitions are parallel in formant space, heading towards the same
formant target, in word-final open syllables, the second mora target can be undershot
(and in this context, only the long variant occurs). Let us turn now to word-internal /ai/,
which provides evidence to support this analysis.
21.1.5
Case study A. Part 2: Cross-speaker minimal pairs
The phonetic nature of any contrast will vary, so should be examined in a range of
different environments. Unfortunately there are no minimal pairs and no near-minimal
pairs which could provide experimental materials to investigate quasi-phonemic
contrast in /ai/ in any context other than tautomorphemic /d/ vs. suffix /#d/, because
elsewhere, long and short variants are allophonically conditioned.
However, our materials included some trochaic words with /ai/ in the initial
stressed syllable, e.g. sidle and bible, and so it is possible to examine near-minimal pairs
on a speaker-internal basis (e.g. if someone has a long libel vs. a short bible), but this
introduces coarticulatory confounds. However, we repeatedly found individual
differences in the lexical incidence of short vs. long /ai/ among young female speakers
for a single word, e.g. bible (Scobbie and Stuart-Smith, 2008).We realised we could
therefore extend the fundamental concept of the minimal pair test, examining
structurally-matched pairs across speakers. This methodological innovation is a
powerful tool for the controlled experimental study of phenomena which cannot be
examined within a single speaker’s grammar. A word like bible, for example, must have
either a short /ai/ or a long /ai/, because interspeaker variation is largely bimodal.1 The
cross-speaker minimal pair lets us examine the phonetic properties of both /ai/ variants
in a consistent and novel distribution (in this case [b__bəl]) which does not exist in the
grammar of any individual (because the phonetic difference does not condition a
difference in lexical meaning). Free variation within an otherwise relatively
homogenous accent group, or class-based variation, is most useful, since both avoid
phonetic effects of age or sex.
Applying this technique, we found that vowel duration was not representative of
the long/short variant difference (Scobbie, Turk and Hewlett, 1999). Figure 3a shows
that within words, duration of /ai/ is unlikely to be a robust cue to the category. These
young females from both social groups have, for comparison, normal duration
differences in the quasi-phonemic context, where /ai/ before word-final /t/ and /d/ is
198ms and 204ms (n.s.) but before /#d/ is 230ms (where /d/ vs. /#d/ is significant in a
paired samples t-test, t(8)=1.89, p<0.05), and where crisis (short and/or long) vs. miser
(182ms) also has a clear duration difference.2
1
Whether intraspeaker variation is gradient or categorical is a topic for future research.
New footnote: we mean that the longer duration vowel in miser is due to the following /z/ rather than the
/s/ in crisis, where the vowel has a shorter duration, whether or not it is phonologically long or short.
2
F2 (Hz)
2750
175
2250
1250
250
150
350
125
100
450
75
550
50
650
25
0
miser (0/7)
crisis (6/8)
libel (0/4)
pylon (0/8)
hydro (0/8)
nitro (3/7)
sidle (6/7)
micro (4/8)
long
750
bible (6/8)
short
a.
1750
F1 (Hz)
duration of vowel (ms)
200
short
long
850
950
b.
Figure 3 a. Mean duration of /ai/ in trochees, separated by impressionistic categorisation of vowel into
long vs. short variants, for 8 young female Glaswegians. The number of short tokens and the total number
of tokens are indicated in brackets. b. Mean formant transitions of the two variants of /ai/ in crisis with
marks at 25ms intervals from the same speakers.
Figure 3b (for crisis) suggests that transition timing (with first mora undershoot)
is again a strong correlate of the short/long distinction, while overall duration is far less
important (125ms vs. 139ms for crisis in Figure 3a), or irrelevant. Basically, short /ai/
has early transitions (within 25ms of the start of the approx 125ms vowel) whereas long
/ai/ begins its transition after a delay of about 50ms. Comparable findings from other
words and other speaker groups support this conclusion. For example, among the older
males, all four WC use a short variant in crisis, and all four MC the long one (Figure 4).
F2 (Hz)
2500
2000
1500
1000
250
350
F1 (Hz)
450
550
short
long
650
750
Figure 4 Short (all WC) and long (all MC) socially-meaningful variants of crisis by eight older male
Glasgow speakers.
The lack of an overall durational difference is probably caused by undershoot due to
durational compression, and perhaps different speech rates could be used to examine
this more closely. However, the cross-speaker method is clear and generally applicable.
21.1.6 Sparsely populated phonotactic cells and low frequency items
Lexical frequency is an important factor in much laboratory-based research, and
extending this work on /ai/, for example, means looking at rare words (e.g. sisal, taigon,
Krug, Beeb, oblige) (Scobbie, 2005). This turns a simple fact into a problem: people do
not have the same lexicon, nor are frequency counts necessarily meaningful for unusual
words (see Jurafsky 2003 for more discussion), and elicitation is hard other than
through reading. In the sociolinguistics context, researchers have experience of the
performance problems associated with reading aloud from text while being observed by
pronunciation researchers, the disparity between oral vernacular lexical items and
standard written ones (but see Macaulay 1991), and cross-dialectal borrowing.
Reading skills become an issue for even high frequency items in simple
sentences, so appropriate methods may be required (see below.) One approach, piloted
in Scobbie (2005), is to use carrier sentences which include the meaning of the relevant
word and ask the participant for their estimated age of acquisition after reading the
sentence twice (with an additional single citation form). This draws attention away from
pronunciation and provides useful information on whether the item is known or not.
Another alternative we have piloted is to use semantic sets in the elicitation of single
citation forms, to focus attention away from phonological relationships. To investigate
Scottish /w/ vs. /ʍ/, for example, one set might be “Scotland, Ireland, Wales, England”
and another “dolphins, turtles, whales, fish”. For efficiency, sets should probe multiple
topics.3
21.1.7 Case study B: vernacular articulations in the laboratory
For technical quality and for logistical reasons it is probably preferable to collect
vernacular data in the laboratory, and despite the obvious difficulties in relaxing the
participants, it is possible to collect good casual/vernacular speech data in this setting
(see Warner, this chapter, Anderson et al., 1999). However, such research rarely if ever
employs social stratification, and may even fail to record simple social demographic
sociolinguistic information. It may also not be explicitly aiming for vernacular speech.
However, using more intrusive phonetic instrumentation would still seem
genuinely problematic. Another study that we have conducted has investigated the
speech production of a socially-structured sound change in progress, namely
3
New footnote: To some extent re-ordered sets can elicit repetitions of the same item in different listpositions to allow for investigation of phrase-final lengthening or sandhi co-articulation, for example, but
multiple repetitions of the same lexeme tend to increase the chances of speaker awareness.
derhoticisation of coda /r/ (Scobbie, Stuart-Smith and Lawson 2008). This Ultrasound
Tongue Imaging (UTI) study used an ultrasound probe mounted on a stabilising
headset, attached to a processing box, in conjunction with an audio recording, with a
team of experimenters/technicians present. Unsurprisingly, typical sociolinguistic data
elicitation is naturally biased towards the acoustic analysis of speech, because an
anonymous microphone appears to offer the prospect of reducing the Observer’s
Paradox in a straightforward way. Articulatory phonetics, on the other hand, is a priori
more intrusive. Yet it is important to merge these different methodologies to help
provide data that exemplified a wider set of systems and is more ecologically-sound.
We investigated the relative impact of different experimental conditions on fine
grained aspects of vernacular speech performance in young working-class adolescent
males from West Lothian, Scotland. First, pairs of participants were recorded chatting to
each other and reading a wordlist in their school, with and then without the UTI
equipment. No significant differences in consonantal variation were found between the
two conditions (Lawson, Stuart-Smith and Scobbie 2008). Then four of the same
participants were brought to the laboratory, where audio and ultrasound recordings of
read and spontaneous speech were made. Informal impressions are that there are few
differences in the level of vernacular obtained between the field and the laboratory
recordings. This same impression was maintained for the subsequent socially-stratified
corpus (ECB08) which was collected from a further fifteen adolescents. It emerged
from these experiments that physical context, i.e. a university laboratory, mattered less
than the more personal factor of the interlocutor. Having a peer – and friend – present
with a participant seemed to be a strong predictor of more natural speech. This finding
also highlights another aspect of data collection which is often manipulated in
sociolinguistic fieldwork and laboratory phonology alike, namely the use of varied tasks
to provoke variation within speakers. From sociolinguistics, however, we can see
speakers control a range of repertoires from the more to the less vernacular, a systematic
range of linguistic specification. This range can be tapped into by recording more than
one speech style by varying formality, for example (though note that recent studies
show that reading tasks may not always elicit the most standard forms from younger
speakers, Stuart-Smith et al. 2007).
The exciting outcomes of the research include, of course, data on how
vernacular Scottish is articulated. Just as important are the insights that emerge because
the data comprises part of a socially-structured set of linguistic variation. For example,
the complex articulatory-acoustic relationships found in /r/ (Mielke, Baker and
Archangeli, 2010) now have to be seen in the context of the range of rhoticity observed,
from the traditional rhotic Scottish English (which is comparable to North American
English) through to the covert rhotic articulations observed in vernacular Scottish
English (Scobbie, Stuart-Smith and Lawson, 2008). In other words, social factors affect
the much-researched phenomena of intergestural coordination, gestural strength and the
articulatory-acoustics relationship just as much as linguistic ones such as segmental
context or prosodic context.
21.1.8
Conclusions and a recommendation checklist
Laboratory phonology has successfully shown that phonological research must not
restrict itself to introspective study by highly-educated middle class linguists. Our point
is that we must not fall into the trap of undertaking laboratory studies on a similar
narrow group of research study participants. Having rejected introspection, empiricallyminded phonologists (and phoneticians) should be drawn to the lure of vernacular
variation, because purportedly neutral experimental participants exemplify only a small
proportion of the sound system phenomena within any given language, and their
linguistic systems are subject to standardisation and other subconscious or conscious
prescriptive pressures. Moreover, there are methodological and theoretical advantages
in incorporating socially structured pools of participants into experimental design,
whether they include speakers of standard varieties or not, simply because this offers a
fresh dimension of conditioning factors that are relevant to phonology and phonetics.
Socially and geographically structured micro-variationism are extremely useful
additions, we believe, to the experimental toolkit, particularly suitable for exploring the
structuring of systematic fine phonetic detail (e.g. Hawkins and Smith, 2001).
There are many advanced and introductory works on sociolinguistic
methodology, too numerous to list in full here, which should be consulted for insights
into how to address certain core issues, such as defining social class, recruitment, and
the tracking of social networks (e.g. Labov 2001; Chambers, Trudgill and SchillingEstes 2002; Milroy and Gordon 2003; Llamas, Stockwell and Mullany 2006;
Tagliamonte 2006; Meyerhoff 2006). In addition, bilingualism research offers useful
guidelines on encouraging the use of a particular language mode (e.g. Grosjean 2008).
Instead, and to conclude, we offer some methodological topics to consider, which we
think are useful starting points for the standard laboratory phonologist.

Consider your question from a wider perspective – have sociolinguistic studies been
carried out on the variety and/or the feature you are examining?

Consider stratifying your participants on some basis which is likely to provide
structure to interspeaker variation which appears to cross phonological category
boundaries.

Avoid exclusive use of graduate students, but widen your pool to undergraduates,
including from other subject areas, such as engineering/physical sciences.

Brief your participants that natural relaxed speech is the goal and misdirect them by
stressing that the recordings are being made for some purpose other than judgement
of correctness, such as measuring the “noise” of talking, and that they need not aim
to speak properly or carefully.

Experimental participants can often be encouraged to be more vernacular by being
accompanied or even observed by a (quiet) friend from the same background, who is
permitted to sit in the lab to exert subliminal pressure to conform to type. Let the
friendship pair converse spontaneously and unmonitored before experimental
materials are used to let them become accustomed to the lab. Vernacular jokes or
stories may also make a useful ice-breaker. Use a non-native or alternatively a
vernacular research assistant or technician.

Pictures and oral prompts may work better than orthographic prompts, particularly
for vernacular lexical items which are not often written down. Try designs in which
participants can repeat or respond to audio prompts in vernacular or other accents
and not merely to read aloud from standard language.

If you can control for the generally poor acoustic environment, try fieldwork in
schools, shopping centres or museums for mass observation research where a high
number of participants can be recruited in a short time.4

Articulatory instrumentation can be sold to speakers as being a set of tools for
understanding physiology (the size, shape and functioning of the vocal organs),
where speech is just a way of getting the organs to move.

4
Finally, and most importantly: collaborate with sociolinguists.
New footnote: we were taking inspiration from some of our own public outreach work, but also the
methods used by the project “r-kennen, socio-dialectological, phonetic and phonological qualities of /r/ in
Dutch” (Hans Van de Velde and Wim Zonneveld) and the HEMA corpus (2001-2003). See:
Tops, E. (2009) Variatie en Verandering van de /r/ in Vlaanderen. Brussels: VUBPress
Sebregts, K. (2014) The Sociophonetics and Phonology of Dutch r. Utrecht: LOT.
Our final remark is to make a plea for the consideration of varieties of language
which are “outside the box”. We think in particular of so-called “mixed” systems,
arising from multilingualism or multi-dialectalism. Both laboratory phonology and
sociolinguistics find such systems challenging, and yet there is a great deal that can be
learned from such a natural part of typical language acquisition and function. Thus
simpler is not necessarily better. By incorporating what may sometimes be dismissed as
extraneous social factors in addition to the typical phonetic, phonological, and
psycholinguistic ones, laboratory phonology will provide an evidence-base more
powerful for those with narrow interests and more representative for those with broad
ones.
Acknowledgements
Scobbie thanks ESRC (Fellowship R000271195) and Stuart-Smith the Leverhulme
Trust for support for the reported work on /ai/. They thank Eleanor Lawson for
collaboration in the creation of the ECB08 corpus, and acknowledge ESRC (RES 00022-2032) for funding it, and Claire Timmins for collaboration in collecting the Glasgow
data.
References (unchecked / uncorrected)
Allen, J. Sean, Joanne L. Miller and David DeSteno (2003). ‘Individual talker
differences in voice-onset-time’, Journal of the Acoustical Society of America 113, 544–
552.
Anderson, Anne, Bader, Miles, Bard, Ellen Gurman, Boyle, Elizabeth, Doherty,
Gwyneth, Garrod, Simon, Isard, Stephen, Kowtko, Jacqueline, McAllister, Jan, Miller,
Jim, Sotillo, Cathy, Thompson, Henry S. and Weinert, Regina (1991). ‘The HCRC Map
Task Corpus’, Language and Speech 34, 351-366.
Bucholtz, Mary (2003). ‘Sociolinguistic nostalgia and the authentication of identity’,
Journal of Sociolinguistics 7(3), 398-416.
Campbell-Kibler, Kathryn (2007). ‘Accent, (ING), and the social logic of listener
perceptions’, American Speech 82(1), 32-64.
Chambers, Jack, Trudgill, Peter and Schilling-Estes, Natalie (eds.) (2002). The
Handbook of Language Variation and Change. Oxford: Blackwell.
Cho, Taehong and Peter Ladefoged (1999). ‘Variation and universals in VOT: evidence
from 18 languages’, Journal of Phonetics 27, 207–29.
Coupland, Nikolas (2007). Style: Language variation and identity. Cambridge:
Cambridge University Press.
Docherty, Gerard J. (2007). ‘Speech in its natural habitat: Accounting for social factors
in phonetic variability’, in Jennifer Cole and José Ignacio Hualde (eds.), Change in
Phonology (Laboratory Phonology 9). Berlin: Mouton de Gruyter, 1-35.
Docherty, Gerard J. and Foulkes, Paul (2000). ‘Speaker, speech and knowledge of
sounds’, in Noel Burton-Roberts, Philip Carr, and Gerard J. Docherty (eds.),
Phonological Knowledge: Conceptual and Empirical Issues. Oxford: Oxford University
Press, 105-129.
Drager, Katie (in press). ‘Sensitivity to grammatical and sociophonetic variability in
perception’, in Paul Warren and Jannifer Hay (eds.), Laboratory Phonology 11.
Docherty and Mendoza-Denton, this volume.
Foulkes, Paul and Gerard J. Docherty (2006). ‘The social life of phonetics and
phonology’, Journal of Phonetics 34, 409-438.
Gordeeva, Olga B. (2008). ‘The relative importance of laryngeal, supralaryngeal and
temporal speech production levels in the implementation of the Scottish Vowel Length
Rule’, poster paper presented at the Meeting of the British Association of Academic
Phoneticians (BAAP), Sheffield.
Grosjean, François (2008). Studying Bilinguals. Oxford: Oxford University Press.
Hawkins, Sarah and Smith, Rachel (2001). ‘Polysp: a polysystemic, phonetically-rich
approach to speech understanding’, Italian Journal of Linguistics – Rivista di
Linguistica 13, 99-188.
Jurafsky, Dan (2003). ‘Probabilistic Modeling in Psycholinguistics: Linguistic
Comprehension and Production’, in Rens Bod, Jennifer Hay, and Stefanie Jannedy,
(eds.), Probabilistic Linguistics. Cambridge Mass: The MIT Press,39-96.
Labov, William (1972). Sociolinguistic patterns. Philadelphia: Univ. of Pennsylvania
Press.
Labov, William (2001). Principles of Linguistic Change: II Social Factors. Oxford:
Blackwell.
Lawson, Eleanor, Stuart-Smith Jane and Scobbie, James M. (2008). Final Report for
ESRC Research Grant “Looking variation and change in the mouth: developing the
sociolinguistic potential of Ultrasound Tongue Imaging” (RES-000-22-2032). [Available online
at http://www.esrcsocietytoday.ac.uk]
Llamas, Carmen, Mullany, Louise and Stockwell, Peter (eds.) (2006). The Routledge
Companion to Sociolinguistics. London: Routledge.
Macaulay, Ronald (1991). Locating Dialect in Discourse: The language of men and
bonnie lassies in Ayr. Oxford: Oxford University Press.
McKenna, Gordon (1988). Vowel duration in the Standard English of Scotland.
Unpublished M.Litt. Thesis, University of Edinburgh.
Meyerhoff, Miriam. (2006). Introducing Sociolinguistics. London: Routledge.
Mielke, Jeff, Baker, Adam and Archangeli, Diana (2010). ‘Variability and homogeneity
in American English /r/ allophony and /s/ retraction’, in Cécile Fougeron, Barbara
Kühnert, Mariapaola D’Imperio and Nathalie Vallé (eds.), Variation, Phonetic Detail,
and Phonological Representation: LabPhon 10. Berlin: Mouton de Gruyter, 699-729.
Milroy, Lesley (1987). Observing and Analysing Natural Language. Oxford: Blackwell.
Milroy, Lesley and Gordon, Matthew. (2003) Sociolinguistics: Method and
Interpretation (2nd Edition). Oxford: Wiley-Blackwell.
Ohala, John J. and Jeri J. Jaeger (eds.) (1995). Experimental Phonology. Orlando, FL:
Academic Press.
Pierrehumbert, Janet, Beckman, Mary, and Ladd, D. Robert (2000). ‘Conceptual
Foundations of Phonology as a Laboratory Science’, in Noel Burton-Roberts, Philip
Carr, and Gerard J. Docherty (eds.), Phonological Knowledge: Conceptual and
Empirical Issues. Oxford: Oxford University Press, 273-303.
Scobbie, James M. (2005). ‘The “end” of phonology: the theoretical significance of
interface phenomena’, oral paper at the 1st International Conference on the Linguistics
of Contemporary English. University of Edinburgh, Scotland, 23-26 June.
Scobbie, James M. (2006). ‘Flexibility in the face of incompatible English VOT
systems’, in Louis Goldstein, D.H. Whalen and Catherine T. Best (eds.), Laboratory
Phonology 8: Varieties of Phonological Competence. Berlin: Mouton de Gruyter, 367392.
Scobbie, James M. (2007). ‘Biological and social grounding of phonology: variation as
a research tool’, in Jürgen Trouvain and William J. Barry (eds.), Proceedings of the
16th International Congress of Phonetic Sciences, Saarbrücken, 225-228.
Scobbie, James M., Hewlett, Nigel and Turk, Alice E. (1999). ‘Standard English in
Edinburgh and Glasgow: the Scottish vowel length rule revealed’, in Paul Foulkes and
Gerard J. Docherty (eds.), Urban Voices: Accent Studies in the British Isles. London:
Arnold, 230-245.
Scobbie, James M. and Stuart-Smith, Jane (2008). ‘Quasi-phonemic contrast and the
fuzzy inventory: examples from Scottish English’, in Peter Avery, Elan B. Dresher and
Keren Rice (eds.), Contrast: Perception and Acquisition: Selected papers from the
Second International Conference on Contrast in Phonology. Berlin: Mouton de Gruyter,
87-113.
Scobbie, James M., Stuart-Smith, Jane and Lawson, Eleanor (2008). Looking variation
and change in the mouth: developing the sociolinguistic potential of Ultrasound Tongue
Imaging, Final report to ESRC, Research Grant RES-000-22-2032.
Scobbie, James M., Turk, Alice E. and Hewlett, Nigel (1999). ‘Morphemes, phonetics
and lexical items: The case of the Scottish vowel length rule’, Proceedings of the XIVth
International Congress of Phonetic Sciences Volume 2, 1617-1620.
Stuart-Smith, Jane (1999). ‘Glasgow: accent and voice quality’, in Paul Foulkes and
Gerard J. Docherty (eds.), Urban Voices: Accent Studies in the British Isles. London:
Arnold, 203-222.
Stuart-Smith, Jane (2003). ‘The phonology of modern urban Scots’, in John Corbett, J.
Derrick McClure and Jane Stuart-Smith (eds.), The Edinburgh Companion to Scots.
Edinburgh: Edinburgh University Press, 110-137.
Stuart-Smith, Jane (2007). ‘Empirical evidence for gendered speech production: /s/ in
Glaswegian’, in Jennifer Cole and José Ignacio Hualde (eds.), Change in Phonology
(Laboratory Phonology 9). Berlin: Mouton de Gruyter, 65-86.
Stuart-Smith, Jane, Timmins, Claire and Tweedie, Fiona (2007). ‘ “Talkin’ Jockney”?:
Accent change in Glaswegian’. Journal of Sociolinguistics 11, 221-260.
Tagliamonte, Sally (2006). Analysing Sociolinguistic Variation, Cambridge: Cambrige
University Press.
Thomas, Erik (2002). ‘Sociophonetic approaches of speech perception experiments’,
American Speech 77, 115-47.
Trochim, William and Donnelly, James P. (2006). The Research Methods Knowledge
Base, Third Edition. Cincinnati: Atomic Dog Publishing.
Warner, this chapter
Warren and Hay, this chapter
Wassink, Alicia, Beckford, Wright, Richard A. and Franklin, Amber D. (2007) Speaker
variability in vowel production: An investigation of motherese, hyperspeech, and
Lombard speech in Jamaican speakers. Journal of Phonetics 35, 363–379.
Watt, Dominic and Ingham, Catherine (2000). ‘Durational evidence of the Scottish
Vowel Length Rule in Berwick English’, Leeds Working Papers in Linguistics and
Phonetics 8, 205-228.