Analyzing Language 2018

• Time-aligned transcription and audio
• Transcription software
• ELAN, PRAAT

Bear in mind the following when orthographically transcribing:

• Always mark uncharacteristic speech

• Write out numbers in full e.g. three, two hundred and fifty six

• Leave contractions as they are e.g. dunno, not do not know

Units of speech:

• E.g. speaking turns

Why time-aligned transcription?

• Identification of tokens should be based on transcript, not on audio

• Avoid biases

• Principle of Accountability

Discrepancies in (English) spelling

• different spelling → same sound: Phoenix, key

• same spelling → different/no sound: ski, sister, business

  → need to avoid ambiguity

• International Phonetic Alphabet (IPA)

• Each sound represented by one symbol and vice versa

• Standard tool in phonetics/phonology

• Letter symbols (sound)

• Diacritics (sound qualities)

• Suprasegmentals (intonation, stress)

GA (General American)

• educated northern US, no local or regional features

• E.g. <bath> → /bæθ/

RP (Received Pronunciation)

• British Isles (“Oxford English”, “Queen’s English”)

• E.g. <bath> → /bɑθ/

• Automatic time-aligned phonemic transcription

• Based on an orthographic transcription + audio file

• Manually check potential alignment and/or transcription errors

• Manually adjust alignment and/or transcription if necessary

● Follow Principle of Accountability

● Exclude

• Uncharacteristic speech

• Overly emotional speech

• Cut-off tokens

• Tokens affected by background noise

data collection and transcription

○ Additional requirements for phonetic analysis
○ IPA
○  Orthographic transcription
○  Token selection

• A handy model for explaining the processes involved in speech production
• We can use this to map articulatory phonetics onto acoustic phonetics (and vice versa)
• Basic premise: there are two independent parts:

1. The source - the sound produced at the glottis

2. The filter - the shape/constriction in the vocal tract

• The frequency at which the folds open and close determines the fundamental pitch of a sound – (the source determines pitch)

• The filter then determines the quality of the sound - whether it’s a /i/ or a /t/

• The filter (the vocal tract) suppresses and lets through sound energy at certain frequencies, depending on its length and shape

• Sound energy thereby gathers at certain frequencies, in peaks – we call these formants.

•  Formants tell us a lot about the sound’s character/quality

Vowel qualities (monophthongs)

IPA Vowel chart: high - low, front - back, open - close

Vowel formants (monophthongs)

Vowel quality

Vowel formant

height - F1

front/back - F2

(roundedness - F3)

Visualization of vowels (monophthongs)

waveform, spectogram, segments

F3, F2, F1 = formants

Vowel onset/offset:

• Start/end of the first/last cycle of regular, repeating sine wave (waveform)

• Start/end of complete set of dark bands for first 3 formants (spectrogram)

Single point measurements

• Midpoint approach: 50%

• Maximal displacement method: 30% or maximum F1

•  pros: simple, fast

•  con: oversimplification

Multiple point measurements

• Default distance approach (e.g. 25 ms from each edge)

• Maximal displacement approach (e.g. 33% + 66%)

• Proportional distance approach (e.g. 20% + 50% + 80%)

• Interval approach (e.g. every 25 ms)

• pro: movement/dynamic

• con: lots of data points make analysis more complicated, time consuming

Maximal displacement approach

Proportional distance approach

•  open PRAAT

• in PRAAT objects window

  → Open
  → Read from file
  → select sound file (.wav) and textgrid (.textgrid)
  → highlight both files, click View & Edit

• in wave form/spectrogram window
  → go to Formant
  → check Show formants
  → place cursor at point of measurement
  → go to Formant
  → select Formant listings

• determine vowel onset and offset

• measure from onset to offset

Token selection

• Primary stress

• Preferably CVC environment (stops, fricatives, affricates)

• Pay attention to phonological environment/coarticulation effects

What to do with all those measurements?

● Descriptive statistics (plotting)

● Plotting software

○ Excel
○ PRAAT
○ NORM suite 
○ R

● Inferential statistics

• E.g. establishing correlations, estimate change over time, etc.

• Determine statistical significance of findings

• Supplement descriptive statistics

If you want to compare speakers...

• Normalize your (acoustic) data!

• Filter out variation caused by physiological differences while preserving sociolinguistic/dialectal/cross-linguistic differences in vowel quality

• Much like vowels, we can measure the quality of consonants on Praat

• They have standard formations in both the waveform and spectrograms

• These are due to the manner and (albeit less so) the place of articulation of the consonant

Visualization of consonants: Fricatives

• Aperiodic waveform

• Place of articulation determines at which frequencies the energy gathers

Visualization of consonants: Nasals

• Similar to vowels, with less amplitude in the higher frequencies

Visualization of consonants: Plosives

• Spectrograms reflect the articulatory process of closure and release

• Silence, then a high intensity period

Measuring consonants

• Consonants can be analysed in various ways

• Methodology can be determined by particular variable and/or researcher’s preference

Voiced and Voiceless Consonants - VOT

• When we hear a plosive, we determine whether its voiced or voiceless from the Voice Onset Time (VOT)

• This is the time between the release and the beginning of the voicing

• can be analysed in PRAAT

• We can use this in speech perception and dialectological studies

Choice of analysis methodology should take into account the variable, the researcher and extensive up-to-date methodology research

• Very well researched variable in the UK, highly salient and commonly stigmatised 

• Recently: Dahou & Hamlin (2016) - ‘Ow Cockney is Beckham Twenty Years On? An Investigation into H-dropping and T-glottaling’

• They used transcription and alignment software to process their data, but then did the analysis themselves, auditorily.

• Pitch variation across segmental features

• Does not affect meaning of individual words, i.e. not like tone in languages like Mandarin or Navajo

• But does convey semantic content

• Varies across languages, but also dialects, so we can analyse it for social/cultural meaning

• Rising intonation in declarative sentences

• Associated strongly with GA and Southern Hemisphere English varieties such as Australian and NZ English

• “Valley Girl” connotations, but now widely used by both genders and in all regions of the US

Indexicalises dialect on a macro level, but what about the immediate context?...

• Discourse functions? Why would we employ HRT?

• What does it signify to the listener in terms of interactional behaviour?

• End of turn is usually marked by semantic, syntactic and prosodic features

• We use syllable length, pauses and rhythm to understand the speaker is about to finish their utterance

• Content and visual queues are also obviously relied upon

• How people perform turn-taking can vary between and in dialects/speech communities/individuals

• We can use auditory and acoustic methods to measure them

A common set of concerns then in fieldwork across sociolinguistics and discourse studies: carefully defining and defending the object of our investigation, justifying it, addressing potential problems with it.

It’s not as straightforward as it might seem: Social class, age, gender, ethnicity, sexuality, social identity...:

 Top-down versus bottom-up categorisations
 Emic versus Etic categories
 Categories versus constructions
 Intersectionality

 All of what is to occur and what might occur are disclosed to the subject;
 The subject should, at some level, understand the research.

 The subject must be competent to make a rational and mature judgment in agreeing to take part in the research;
 Consent must be given freely, voluntarily and with no coercion.

• The study of certain kinds of informant: children, vulnerable adults, ‘infiltration’,

• The observation of public behaviour: Are we free to view and record any sort of public behaviour?

• The Observer’s Paradox... the act of observing will affect the object being observed

• The subject’s right to know and veto VERSUS the pursuit of knowledge...

“Individual and cultural definitions and expectations of privacy are ambiguous, contested, and changing. People may operate in public spaces but maintain strong perceptions or expectations of privacy. Or, they may acknowledge that the substance of their communication is public, but that the specific context in which it appears implies restrictions on how that information is -- or ought to be -- used by other parties”

Top-down: researcher decides on defining categories / provide a set of limited options,
e.g. Reseracher: "Are you Hindu or Muslim?"
Participant is now (kind of) limited to the answers "Hindu" or "Muslim", which may unconsciously influence him/her.

Bottom-up: partcipants decide on defining categories / researcher does not provide any options,
e.g. Reseracher: "Are you religious?"
Participant can now freely choose an answer.

Emic stands for what is represented, e.g. block years (a 28 y.o. study participant would be 20-30 in block years, which is not accurate)

Etic stand for the actual, physical, e.g. the physical years of a study participant (28 in our example)

the interconnected nature of social categorizations such as race, class, and gender as they apply to a given individual or group, regarded as creating overlapping and interdependent systems of discrimination or disadvantage.

1. Speech as a source of data

2. Surveys as a source of data

3. Corpora as a source of data

4. Data collection in Linguistic Landscape research

- excludes interviewer effect
- useful for exploring how everyday language works (data can be collected by recording events when people routinely come together or from self-recordings)
- ethnographic approaches

..........extended visits among friends and family (Milroy and Milroy (1985)
..........dinnertime narratives (Ochs and Taylor 1995)

..........collecting teenager interactions (The Bergen Corpus of London Teenage Speech (COLT)),
..........collecting style repertoires (Sharma 2011)

..........collecting speech from children (Smith 2003-2005)

•  the researcher is still there as an ‘imagined audience’
• the amount of speech you get from different people will vary
• the settings will vary widely and may not allow for direct comparison
• the researcher has no control over the recordings and the quality may be poor (e.g. background noise, recording volume, type of speech recorded)

• enables the collection of substantial amounts of fluent and relatively natural speech

• a good way of capturing the ‘vernacular’ (the style of speech that people acquire naturally and use at home and with their peers)

• the ‘sociolinguistic interview’  - ‘semi-structured’ and is a good way of collecting different speech styles (‘casual’ style, ‘interview’ style and, when combined with reading tasks, more ‘formal’ styles)

• the ‘informal interview’ (often involving dyads or small friendship groups) - ‘unstructured’ and can trigger the most natural, spontaneous speech and is often used in combination with ethnographic approaches 

• time-consuming (for interviewer and interviewee)

• the interview may not ‘flow’ well and the interaction only elicits minimal responses

• the ‘observer’s paradox’ 
  -> ‘To obtain the data that is most important for linguistic theory, we have to observe how people speak when they are not being observed’

• Never begin an interview cold. Spend a few minutes ‘breaking the ice’.

• Remember your purpose. Make sure you get answers to essential

  questions.

• Don’t be satisfied with monosyllabic answers. Avoid ’yes/no’ questions and try ‘probing’ with further questions e.g. ’what happened then’?

• Demonstrate aware hearing. Remain interested and offer non-verbal responses.

• Practice, practice, practice! Go out and do some interviews and reflect on what went well and what didn’t go so well.

• Eckert spent 2 years in a Detroit high school, observing and recording participants from one graduating class

• She identified different social categories within the school and was able to analyse language use among the different groups. Getting to know the students over a period of time enabled her to conduct informal interviews as well as gather information through her observations

• She found differences in language use among the Jocks (those whose lives centred around school activities) and the Burnouts (those who were least engaged at school and tended to leave school to go straight into the workplace).

• useful for accessing information on linguistic behaviour that is not easy to observe or record

• useful for greater breadth of coverage, in terms of population and number of features examined