Handout #1Data in the study of speechExperiments• An experiment is a research study with thefollowing properties:– (1) The data consist of objectivemeasurements.– (2) There is a comparison of two or moregroups of measurements.– (3) There is control of confounding factors.Objective measurements• A measurement is a number that represents anevent or object.• A measurement is objective if you can define aprecise procedure that would allow anotherresearcher to obtain the same number for the sameevent.• An objective measurement does not vary fromobserver to observer, as long as the definedprocedure is followed.Objective measurements inspeech• Acoustic measurements: duration, fundamentalfrequency, formant frequency, intensity….• Physiological measurements: glottal width,tongue tip displacement, …• Aerodynamic measurements: pressure, volumevelocity,…• Perceptual measurements: frequency of a givenresponse, response time, …Advantages of objectivemeasurements• They are numbers, so we can usequantitative analysis techniques todetermine how reliable our generalizationsare.• They are not distorted and obscured bysubjective biases of the observer.• They are fine-grained enough to capturegradient generalizations.Judgements as data• The data presented in most linguistic work onspeech sounds consists of transcriptions: anobserver’s judgements as to what sequence ofsound categories occurred.• This has been the case up to now in phonology,historical linguistics, sociolinguistics, andlanguage acquisition.• It remains the norm in linguistic work on speechin journals and conferences, except for phonetics.Judgements as data• Most work in syntax is based ongrammaticality judgements by nativespeakers.• Semanticists work often with similarjudgements of whether an utterance isanomalous.Problems with judgement data• There are some inherent problems with suchjudgements:– They are subjective.– They cannot represent gradient patterns.– They are metalinguistic rather than linguistic.Subjectivity of transcriptions• Transcriptions represent a subjective judgement ofthe transcriber.• They vary from transcriber to transcriber, evenwhen they are transcribing the same material.• In Kinyarwanda, for example, there are 5 differenttone transcriptions for a word like word umusore“youth” in 5 different publications (Myers 2003:73).Subjectivity of transcriptions• Some other cases of intertranscriber variation:– Secondary stresses– Vowels in American English accents (Labov1994)– Final high tones in Chichewa (Myers 1999)• Grammaticality judgements also vary greatlyamong members of the same speech community.Bias• The variability in judgements is not random, but isshaped by the observer’s expectations.• The transcriber’s linguistic background inevitablyplays a part in the categories he/she hears.• Grammaticality judgements change over the years,in sync with the changes in the predictions ofsyntactic theories.Bias• Labov (1994) discusses cases in whichtranscriptions of American English dialects weredistorted by transcriber biases.• Transcriptions in the dialectological literaturehave frequently been normalized, eliminatinglocal variants.• Cases of near merger, where two categoriescannot be reliably distinguished by nativespeakers, but remain acoustically distinct, havebeen rendered as complete mergers.Subjectivity• Our goal is to find out about language andlanguage users, not about the transcriber.• Due to their subjectivity, transcription datais not replicable.• If we have doubts about a generalizationbased on transcription data, we can’t test thegeneralization by redoing the transcription.The categoriality of transcriptions• Transcriptions are inherently categorical -the transcriber assigns continuous soundevents to categories.• But many sound patterns are demonstrablygradient, and so inexpressible in such arepresentation.Nasal vowels in English• The classic description of nasal vowels inEnglish is that they occur only beforenasals, and elsewhere only oral vowelsoccur.• Cohn (1993) used airflow measurements toshow that there is greater nasal airflow thecloser one gets to a nasal on either side.Nasal airflow in deed and deanNasal vowels in English• It would be inaccurate to say that the vowelin dean is nasal.• It is continuously more nasal over thecourse of its whole duration.• This is a gradient coarticulation effect,which would be distorted by a transcriptionof that vowel as nasal.Gradience• It turns out that a lot of sound patterns aregradient, e.g. Sproat and Fujimura (1993), Choi(1995), Myers (1999, 2003).• Labov (1994) has demonstrated that thedifferences in speech among different socialgroups are inherently gradient (cf. Labov, Yaegerand Steiner 1972).• The process of language acquisition is gradient:Menyuk and Klatt (1975), Smith (1979), Mackenand Barton (1980).Gradient patterns• Gradient patterns can only be revealed as such infine-grained continuous data.• In a high proportion of the cases in which suchdata has been explored, patterns that had beendescribed as categorical have turned out to begradient.• A transcriber faced with a gradient pattern musteither ignore it, or distort it by representing it ascategorical.Metalinguistic judgements• Even when the transcriber is a nativespeaker of the language being described, thejudgements do not directly reflect theunconscious automatic categorizationjudgements that are part of his/her nativespeaker knowledge.• Transcribers are conscious metalinguisticjudgements.Metalinguistic judgements• All native speakers know the categories oftheir language.• But not all native speakers are competent atphonetic transcription.• As Sapir (1933) showed, native speakerstend to hear only the phonemes of theirlanguage, glossing over even the mostsalient allophones.Metalinguistic judgements• Talking about language is quite different thanspeaking a language.• Native speakers do not have any access to theirunconscious knowledge.• If they had access, they could write a grammar byjust writing that knowledge down.• Transcriptions represent not the categories of thelanguage, but the categories of the describer’sanalysis of the language.Laboratory phonology• Laboratory phonology is the use ofexperimental method to address issues inphonology.• It is a movement attempting to replacesubjective data with objectivemeasurements.• There is a biennial