Tags:
Computing Linguistics Paperville psychology statistics Bob Port lists many interesting papers on this page in support of Rich Phonology, which I know little other than that it seems to emphasize on the richness (high dimensionality) of memory/representation of spken words (or do words exist if we are anti-segmental all the way?).
I am far from qualified to comment on that. Instead, my interest is in the school of thought that the intuition of phonemes/segments comes from alphabetic writing; though I did digress at the end.
I was attracted to the following papers:
Port, Robert (2006) The graphical basis of phones and phonemes. In Murray Munro and Ocke-Schwen Bohn (eds.) Second Language Speech Learning: The Role of Language Experience in Speech Perception and Production. Benjamins, Amsterdam. pp. 349-365.
[Update: I really like this paper and agree with the central notion that segmental representations are just a tool. That’s what I said earlier (below), but Port argued more forcefully. Not being a linguist or a phonetician, my concern was less on what kind of a theory is necessary to describe language, but more on how alphabetic/segmental writing is possible. To reiterate what I was about to say later, the mapping from the [unknown, possibly hi-D] psychologically real representation of speech to a segmental representation cannot be terribly difficult. To this end I find OT useful, whereas Bob Port’s advice to graduate students was basically not to touch it.]
Olson, David R. (1993) How writing represents speech. Language and Communication 13. 1-17. This article is a revised version of Chapter 4 in his outstanding and highly recommended book: The World on Paper: The Conceptual and Cognitive Implications of Writing and Reading. (Cambridge: 1994). He points out that various human notions of what language is are invariably based on whatever we represent with our writing system. So the idea that our alphabetical writing is a `cipher’ (one-to-one replacement) from phonemes into graphemes has reversed the true order! We linguists tend to believe our language is made from phonemes primarily because we use letters to represent speech.
Faber, Alice (1993) Phonemic segmentation as epiphenomenon: Evidence from the history of alphabetic writing. In Downing, Lima and Noonan (eds) The Linguistics of Literacy. (Benjamins,Amsterdam) pp. 111-134. Faber anticipated the argument I am making here over a decade ago. Her paper has been largely ignored, as far as I can tell. She argued that languages seem to be composed from letter-like segments primarily because we have all been trained in alphabetic literacy.
Abercrombie, David (1949) What is a `letter’? Lingua 2 , reprinted in D. Abercrombie (1965) Studies in Phonetics and Linguistics. (Oxford Univ Press, Oxford), pp. 76-85 A quick, historical look at confusion from the Greeks to modern times about whether letters (Lat. litera) are graphic tokens or `speech sounds.’ Abercrombie thought the story was very clear in 1949 — since, of course, he assumed that phonemes were the correct cognitive units. But it seems to me that he was still very confused himself about these issues. Neither he nor most other phoneticians were ever able to sort out the source of our vivid intuitions about the segmental structure of speech.
Nonetheless, an interesting question remains, a paradox of sort: if segmental representation is "unnatural", how does everybody get it?
Well, the unflattering truth is that not everybody does. Some children figure it out pretty much on their own, while others nevery do. The majority of English-speaking children requires years of instructions and drills.
The twist, however, is that I happen to believe that no amount of drills can make you do something that you are inherently incapable of doing — such as flying or in my case playing the piano. If we can get 70% children to read something as opaque as the English writing system, I am convinced that that is some sort of natural support built-in for segmental representations.
To be clear, I am not arguing that speech sounds ARE segmentally represented in our minds. Rather, whatever mental representation we have, we can fairly easily convert to segmental representations.
As to why most language theories hang on to a segmental representation, very simple: it’s because of the symoblic nature of communications. It’s pretty much the only practical way (so far) to talk about language without showing spectrograms or passing a Python Object. It’s the combined result of the symbolic nature of human language, the requirement for replicability of science, and the unavailability of portable computers capable of visualizing hi-D data in the Greek times. Seriously, humans have rich memory representations, but we have NO way of sharing that rich representation with somebody else. For the sake of science, we had to settle on the cheapo substitute — symbols.
As with connectionist models, the inherent limitation of hi-D models of cognition is not the models themselves, but our ability to communicate what they MEAN. It’s a non-question from a pure engeering/computation point of view: the model does its work; you can try to improve it or throw it away, but often times the most frustrating part of modeling work is to explain to somebody else what certain state of the model "means". That is, you are asked to digitize, discretize, symbolize, verbalize, or segmentalize the hi-D "experience" of the model, and compare that slice of "experience" with somebody else’s model or thoery that has a completely different archetecture and vocabulary.
Pretty hopeless, right? It should also sound familiar, if we imagine each individual speak runs an idiosyncratic hi-D language model in their head. I’d think a lo-D, simplistic theory is a pretty good place to begin a shared understanding. Even with the computing power today, I am not sure we have a solution to the — human — problem of communicating hi-D data/theories.
Coming back to the issue of segmental representation and literacy, there is no question that this is not an easy skill to acquire for some children, even though we have figured out a system to get the majority of kids on the path to literacy. I close with the following question: given the "unnatualness" of segmental representation, should we expect 100% of children to be able to read by 2014, as mandated by No Child Left Behind?
] 
. In fact, you can’t find me even if you turn the "face-only" thing off. My invisibility cloak fooled Google. Horay!] 
. NPR provides an interesting API for accessing their stories; this wedget is one that uses the field:date to plot a time line: 
November 20th, 2008 at 10:02 am e
I videotaped some reading classes around Michigan last Spring, and was shocked to find that “Reading First” means they spend 90 minutes straight first thing in the morning doing reading (even first graders who can’t read). So I’m not surprised to find that they’ve replicated the results about massed vs. distributed practice.