A story about how 'I pahked my cah in Hahvahd Yahd'

Eric Raimy
Information Working Group
June 8, 2004

"It is important to emphasize again that the way that sensory inputs are grouped by our nervous systems determines the patterns that we perceive"

Albert S. Bregman Auditory Scene Analysis

Goal is to present a case study using Grobstein's "Bipartite Brain Model" as applied to linguistic sound structures

1.0 Grobstein's Bipartite Brain Model (1) Grobstein's Revised "Bipartite Brain" Model Active Decoder Model Story Inanimate Builder Teller 2.0 The modular organization of linguistic grammars (2) brain Social Module ear Auditory Phonetic Phonetic Phonology Processing Universal Lang Spec * Roles of each of the modules: Ear- biological structure that determines aspects of what we hear e.g. 20-20,000hz response Auditory Processing- biologically determined processing of auditory stimuli e.g. 'streaming' (Bregman 1990) of auditory stimuli, DECODER Phonetic Universal- biologically determined processing that identifies 'language' and begins language specific processing, DECODER Phonetic Language Specific- emergent module (guided by biologically determined biases) that encodes specific phonetic characteristics defined on a language by language basis MODEL BUILDER Phonology- emergent module (guided by biologically determined biases) that encodes language specific symbolic sound patterns, MODEL BUILDER Social Module- emergent module that encodes social aspects such as SEC, beauty/ugly, smart/stupid, race, gender, etc., STORY TELLER * Evidence for higher level modules: Phonetic Universal: neonates discriminate between language and non-language from birth Phonetic Language Specific: judgments on different dialects, L1 'foreign accents' Phonology: different languages, L2 effects * 'Information' becomes more abstract as higher level processing occurs: Auditory Processing- raw frequency analysis, periodicity, 'timber', etc. Phonetic Universal- only phonetically relevant analyses, formants, transitions, etc. Phonetic Language Specific- abstract phonetic categories Phonology- distinctive features * Lower level 'information' is discarded at higher levels: Phonology - accommodates dialectical differences, difference in male/female/child vocal tracts, etc. Phonetics Language Specific- accommodates 'intradialectical' variance (VOT, different formant values for vowels), acoustic differences e.g. 'high tone' for male may actually be lower in frequency than 'low tone' for female in a tone language (Mandarin) Phonetics Universal- accommodates computer generated language vs. human language * The modules here fall into 'decoder', 'model builder' and 'story teller' depending on the following characteristics -'decoder'- biologically determined and relatively fixed in behavior -'model builder'- guided by biologically determined biases which produces flexible but constrained behavior -'story teller' - input from multiple cross-modal modules, biologically guided (!?) to produce higher level patterns from distinct unrelated patterns * Work accomplished by each module is reduction of variability and enhancement of the stability of the pattern (analog to digital conversion) * CITI Bank Fraud Protection Commercials- the 'story teller at work' -mismatch between appearance and language -effect is only obtainable if we have 'patterns' of relationships between 'appearance' and 'language' -if we separate the visual and auditory components of the commercial, there is no effect 3.0 'r-dropping' in Eastern Massachusetts English * 'r-dropping' is a famous characteristic of dialects of English spoken in the Northeast (among other places) * (3) presents data from the Eastern Massachusetts dialect of English (3) 'r-dropping' in Eastern Massachusetts English (Halle and Idsardi 1997:332) a. the spa[r] is broken the spar is broken b. the spa seems broken the spa[ø] c. algebra[r] is difficult Homer is difficult d. algebra bores me Home[ø] bores me e. the study of algebra the study of Home[ø] * There is also an 'r-insertion' process active in this dialect which produces the homophones seen in (3) * /r/s remain in the memorized forms of words based on the lack of 'r-insertion' in certain contexts (4) underlying /r/ algebra ~ algebr[ej]-ic *algebr[er]-ic Homer ~ Hom[er]-ic *Hom[ejø]-ic * 'r-insertion' only occurs in dialects of English that have 'r-dropping' and the question I want to focus on is why is this the case? * Halle and Idsardi (1997) provide formal analyses of 'r-insertion' and 'r-deletion' according to the rules in (4) (4) a. /r/ deletion X _ Ø / Rime r Nuc X __ b. /r/ insertion Ø _ r / Rime Rime Nuc Nuc [-high] __ X * Halle and Idsardi draw attention to the important fact that the structural description of (4b) contains the structural description of (4a) which has important ramifications according to the 'Elsewhere Principle' (i.e. more specific instructions precede and block less specific instructions) * Halle and Idsardi further suggest that the origin of the 'r-insertion rule' is hypercorrection * Hypercorrection is a sociolinguistic term that refers to when an individual or group changes their behavior to model it after another group's linguistic behavior -perfect imitation is never achieved (post critical period), the hypercorrecting group generally 'overshoots' * 'r-insertion' achieves the goal of 'more /r/s' which is socially desirable but the rule never-the-less still marks the hypercorrecting group as 'outsiders' 4.0 Why we need a 'story teller' for hypercorrection to work * A 'story teller' module is the exact type of entity that we want to explain hypercorrection phenomena * Phonology itself is insufficient because one of its main tasks is to extrapolate away the particular type of information that is necessary * Phonology itself is impervious to 'social information', i.e. no phonological process is sensitive to: male vs. female child vs. adult black vs. white vs. Chicano SEC distinctions * All of the above distinctions are calculated at some level because of judgments people have about these social categories * The speaker who 'hypercorrects' must notice at least the following things (1) they speak differently from 'others' (2) a particular 'linguistic marker' that is socially significant and associated with the 'others' (3) how to use this marker to become like the 'others' * Specifically, for a speaker with an 'r-dropping' dialect: (1) they notice 'others' speak differently (2) the 'others' have more /r/s (3) add more /r/s to my speech * The interaction between the 'story teller' and 'model builder' can be investigated by considering how the behavior is altered * Phonology as a module is 'fixed' at a point in time during the acquisition process where its rules and representations are impervious to change * The 'story teller' module can thus only manipulate the phonology in limited ways: -attenuation of a rule- surface effect is 'variable rules' which appear to have percentages of application on them (i.e. 'r-drop' 45% of the time) -add a more specific rule * Both types of effects are seen in 'hypercorrection' -attenuation of 'r-dropping' can be seen in style shifting -new rule is 'r-insertion' 5.0 Caveats * The story here is only about the genesis of 'r-insertion'- once it has occurred in the dialect, children will learn both rules without recourse to the 'story teller' * All modules discussed here are drastically simplified- each module will have multiple representations and processes within them * Open question about other ways 'story tellers' can interact with 'model builders'-provided evidence about how the 'story teller' modifies phonology but what about other modules, 'vision' References Bregman, Albert S. (1990) Auditory Scene Analysis MIT Press. Halle, Morris and William J. Idsardi (1997) r, hypercorrection and the Elsewhere Condition. In Roca, Iggy (ed) Constraints and Derivations in Phonology. Oxford University Press.