Science Building, Room 227
For further information contact Paul Grobstein.
Summary by Paul Grobstein
Eric outlined an approach to a specific problem of linguistic phonology that drew from earlier conversations in this group and elsewhere about decoders / model builders/story tellers and overall brain organization. In so doing, he provided both a illustrative example of the potential usefulness of a broader conceptual perspective origining outside linguistics and highlighted features of that perspective needing further development. This in turn raised a number of significant issues about information needing further exploration.
Eric proposed for phonology a sequential information processing sequence of steps beginning with decoders and ending with model builders. The distinction between decoders and model builders is a new one, and corresponds to the absence or presence of a capacity for modification reflecting the activity of the processers (roughly a distinction between "hard wired" ("genetic", "biological") and not "hard wired" (notice that the latter is NOT equivalent to "cultural" since modification may or may not reflect "cultural" information and modification always occurs in a structure itself based on genetic information).
Additionally important about Eric's proposed sequence, and the subject of considerable discussion, is the idea that it involves successive "discarding"/ "compression"/"elimination"/"filtering" of information. Eric's point was that the successive modules were achieving increasingly specific information processing tasks and these in turn required ("are defined by"?) "abstraction", or getting rid of information to achieve useful categories for particular purposes.
There was active consideration of the ways in which the variety of terms were similar or dissimilar in meaning. "Compression" for example may be used simply to store X amount of information in a small space, and is done with the presumption that the information removed is not lost but rather recoverable (using an appropriate decoder). "Filtering" on the other hand may involve intentional discarding of information (as in the cases of a lateral inhibition network or a threshholding element such as a neuron). There may also be an important distinction between discarding "noise" and discarding "pattern", both of which occur in various contexts. Finally, it was noted that information that seemed to be "thrown away" (by particular processes in the ns, for example) might in fact be stored/represented elsewhere, and that there may be some relation between this set of issues and the concepts of "filter" and "rectifier" used by the philosopher Michel Serres.
(at summarizer's discretion: the concept of compression/expansion cycles played a big role in a course on evolution and stories that Anne Dalke and I did last semester. Interestingly, in that context compression was regarded primarily not as a process of information reduction but rather as a necessary prelude to expansion, with the couplet being a mechanism for generation of new information ... a subject to be returned to).
The ear and an initial general auditory processing module, Erik argued, are decoders that sample pressure variations using fixed routines. In so doing, they clearly ignore some information (such as high frequencies). They also result in information "streaming", ie the parcellation of different aspects of the incoming signal into separate channels/representations (an analogous process occurs in the retina as a result of different photoreceptors and different classes of ganglion cells). There is a prelude to the model builder/story teller distinction here in that the story teller can, in general, "pay attention" to one or another stream but not all streams simultaneously (yet can make use of multiple streams?).
One additional of the sequential modules is concerned with phonetic universals, ie with the extraction of aspects of the signal received from the auditory processor that relate to language. The argument for this as a decoder rather than a model builder relates to the very early age at which babies seem to discriminate between language and other sounds streams.
Subsequent modules in Eric's scheme are "phonetic language specific" and "phonology" with the former extracting sounds patterns that are specific to particular languages and the latter extracting characteristics that represent language specific symoblic sound patterns. Here too there seems to be information "loss", in two senses. One is the further progressive reduction of the signal to only its task specific meaningful components. A second is a developmental loss of the ability to discriminate particular sounds based on cultural experience.
(at summarizer's discretion: there is a very interesting, very general issue here: there is no doubt but that the extracted signals lack some information present in the original signal (are "filtered", and by removal of pattern as well as noise). Whether the processed (maybe better than "extracted") signals are of lower "information content" is less easily answered. Arguably the decoders (of which model builders are a subset) have ADDED information to the signals in the course of transforming them ... hence offsetting the loss of information content or even enhancing information content? In the visual arena, the lateral inhibition network "adds" the information "edges are meaningful patterns"?)
The "punch line" of Eric's scheme was the idea that the sequence of modules leading up to phonology was not enough to account for some observed phonological patterns, and hence that a "story teller" module was additionally necessary. The argument here depends on the idea that the output of the "phonology" module is necessarily indifferent to aspects of speech that vary among social groups (these have been "filtered out") but that aspects of phonology make most sense on the assumption that individuals are not only aware of phonological variations among social classes but adjust their own phonolical behavior in light of them. This observation, Eric suggested, requires a distinct "social" module that could be understood to be part of the expression of a "story teller".
The relevant specific observations relate to the pattern of "r-dropping" and "r-insertion" in the dialect of English common in Eastern Massachusetts, and to observations on other "r-dropping" English dialects. "r-insertion" occurs only in English dialects which also exhibit "r-dropping" and "r-dropping" is recognized as indicative of lower class status by speakers in general. These observations, together with specifics of the "r-insertion" pattern suggest that r-insertion is a "hypercorrection" originating historically in a wish to appear to be of higher social class.
Beyond the useful notion that the story-teller (consciousness) can modify the organization of a model (as aspect of the unconscious), Eric's story of a story-teller role in phonology has several additional interesting features:
The discussion ended with some quite interesting and relevant meta-level issues. One challenge to Eric's story was in terms of whether it was in fact well-founded in terms of observations (were there or were there not certain observations which the story seems to imply, ie was the story simply summarizing existing information or substantially going beyond it (adding information)?). A second challenge was posed in terms of the question of whether there exist other equally good stories (ie stories that would also effectively summarize (compress?) the existing observations/information). These and related issues warrant further consideration.
| Calendar | About
| Getting Involved
| Groups | Initiatives | Bryn Mawr Home | Serendip Home
Director: Liz McCormack - firstname.lastname@example.org | Faculty Steering Committee | Secretary: Lisa Kolonay
© 1994- , by Center for Science in Society, Bryn Mawr College and Serendip
Last Modified: Thursday, 10-Jun-2004 13:51:24 EDT