Path: utzoo!censor!geac!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!bcm!dimacs.rutgers.edu!aramis.rutgers.edu!paul.rutgers.edu!njin!princeton!pucc!PSYCH@TCSVM From: harnad@phoenix.Princeton.EDU (Stevan Harnad) Newsgroups: sci.psychology.digest Subject: PSYCOLOQUY V1 #16 (Discussion: General - 240 lines) Message-ID: <9012050027.AA05249@reason.Princeton.EDU> Date: 4 Dec 90 23:26:07 GMT Sender: VMNNPOST@pucc.Princeton.EDU (Listserv to Netnews Gateway) Organization: Listserv to Netnews Gateway at pucc.Princeton.EDU Lines: 235 Approved: PSYCH@TCSVM PSYCOLOQUY Tue, 4 Dec 90 Volume 1 : Issue 16 Access to the phonological lexicon (S. Cassidy) Query: Research Organizations (A. Kendall) Response to D.S. Stodolsky's "Consensus Journals" (G. Becker) ---------------------------------------------------------------------- From: steve@comp.vuw.ac.nz (Steve Cassidy) Subject: Access to the phonological lexicon A question about access to the phonological lexicon. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ As a preliminary to my question let me describe the context in which I ask it. I am involved in building a computational model of the development of word recognition. In outline, my model is similar to that described by Seymour et al. (1989): the child starts reading using a purely visual strategy (Seymour calls this Logographic) and later develops a strategy involving recoding words with letter to sounds rules and recognising them via the phonological (speaking/listening) lexicon. My approach is to try and show how each process used within the various strategies can develop from the child's existing capabilities, within the environment in which the child learns. Thus my model of visual recognition is such that I can argue that it can develop from object recognition processes and a growing knowledge of the shapes of letters. As the model acquires more vocabulary items, it can learn more about how best to structure its lexicon for reading. The model as it stands is described in Cassidy (1990a, 1990b). I am now thinking about how the reader starts to use phonological information. The standard account says that the written form is re-coded using some grapheme to phoneme procedure (analogy or non-lexical rules are examples), the resulting phonological form is then recognised using auditory word recognition procedures. The use of non-lexical rules begs the question as to where they came from. One possible solution is that the child can using letter sounds or letter names as proto-rules to do this re-coding. If this is the case then the resulting phonological form will more than likely be incorrect, but it may have about the right consonants in the right places. Analogy might offer a better candidate phonological form but is a very ill defined procedure; I do have ideas as to how it could work but the suffer similar problems of vowel ambiguity. My question concerns access to the phonological lexicon; that is, finding a candidate phonological lexical item given a partial phonological form. How much information is needed and how immune to variation is the access procedure? Which parts of the phonological representation are important -- vowels, consonants, onset, rime? Is there some particular structure, such as onset+rime, which will facilitate access? Is there any evidence that children (or adults) are able to access the lexicon given incomplete descriptions of the target word? If so, does this give any insight into the above problem? Tasks such as rhyme production and alliteration might be suggested but I can imagine another process which could achieve these tasks in a 'generate and test' manner that doesn't require access from partial descriptions. Models of spoken word recognition such as the Cohort model and the Logogen model have very little to say about the information content and structure of the phonological lexicon. In fact they see the lexicon as a collection of processing units, one per word, which sum features of the stimulus until the word is recognised. This could be seen to be equivalent to a lexicon full of descriptions of words which are matched by some process or set of processes against the stimulus. These models don't tell us much about what is in the lexical descriptions, rather they concentrate on how information is collected and matched and how `recognition units' compete for attention. The cohort model might be viewed in this way: the lexicon is a collection of word descriptions indexed under a number of features. When a stimulus arrives the first features available (be they phonetic features, phonemes syllables or whatever) are used to select a subset of the lexicon (those words indexed on that feature), as more features become available they are used to further reduce the size of the subset until there is only one candidate which can then be responded. Under this view it would be possible to use this structure to access lexical items from incomplete phonological descriptions, such as those obtained from a crude letter to sound translation. Seymour, P.H.K. and Evans, H.M. and Kinnison, S.E.C. (1989) Logographic, alphabetic and orthographic processes in early reading development. In Press, Department of Psychology, The University, Dundee, Scotland. Cassidy, S. (1990a). Early Reading Development: A computational view. Technical Report CS-TR-90/8, Dept. Computer Science, Victoria University, Wellington, New Zealand Cassidy, S. (1990b). Substitution Errors in a computer model of early reading. Paper presented to the First Australasian Cognitive Science Conference, Sydney, Australia. Steve Cassidy ========================================================== Computer Science, steve@comp.vuw.ac.nz Victoria University of Wellington, ==================== Box 600, Voice: +64 4 715328 Wellington, New Zealand. Fax: +64 4 712070 ---------------------------------------------------------- ------------------------------ From: "Arthur Kendall" Subject: Query: Research Organizations As part of an effort to consider total-quality management (TQM) approaches in our agency, I am trying to identify organizations whose work is (research, policy studies, evaluation, "think tank", etc.) where there is an explicit management approach that is (humane, nonauthoritarian, professional, total-quality, democratic, participative). If you know of any such organizations, please contact me. If you know people who might have information, please forward this note to them. _________________________________________________________ Arthur J. Kendall National Security & International Affairs Division United States General Accounting Office (GAO) Washington, DC 20548 USA Commercial phone: (202) 275-8455 FTS: 275-8455 BITNET: AJK@NIHCU GAO is an oversight agency in the legislative branch. It is not the same as GSA which is an executive branch agency. *** DISCLAIMER **** This is not an official communication. Only official communications represent official findings, results, or opinions. Any opinions expressed are solely those of the sender as an individual. ------------------------------ From: "Gordon Becker" Subject: Response to D.S. Stodolsky's "Consensus Journals" The statistical procedure that Stodolsky proposes for evaluating articles and for selecting new authors would make it even more difficult than it is now to publish articles that deviate from the dominant paradigm. Current peer-review practices have been highly criticized for their rejection of new ideas. Consensus assures conservatism and Stodolsky's procedure for selecting authors and papers would give technologists conducting minor mop- up experiments more power than scientists breaking new ground. Some of us fear that we are already stifling science with the present old-boy review. We need some way to break out of conservatism, not perpetuate it, and certainly not strengthen it. Stodolsky's method also perpetuates the use of anonymity with all of its disadvantages. Not only does anonymity permit the reviewer to make false and unsupported statements, to be unnecessarily disparaging, personal and disrespectful, all with impunity; it also prevents opening or continuing a fruitful exchange and cooperative effort with helpful reviewers. The claim that reviews would be less critical and candid without anonymity, have been shown to be untrue in the many real life situations where open review has been used. The consensus journal also perpetuates the current practice of considering publication itself as an index of one's "scientific contribution" rather than considering the actual impact of the publication on science. A much better index of one's contribution is the citation index. Perhaps we should let articles remain in the electronic media until they are cited in other electronic articles sufficiently to justify publishing them in hard copy for others not actively engaged in the particular specialty of the article. Thus electronic journals would be for scientists actively engaged in the specialized area of the article and published journals would be for those in related or other areas. A basic problem with any statistical procedure such as that advocated by Stodolsky to evaluate new articles is the need for his assumption of a steady state condition. Research is designed to change the state of knowledge, to generate new ways of organizing data. Significant break-throughs deviate from the past and are generally rejected by the majority when first presented. Any evaluation method that does not consider changes in state, and that gives equal votes to all reviewers is likely to reject the really significant paper. Let's make it easier for anyone to publish a paper in the electronic media, encourage open review, and give more weight to the actual use of the article. The problem for reviewers then becomes one of verifying that the citations are actually justified and relevant rather than inflated by friends and cohorts. It might be necessary to limit the number of citations permitted in an article. The task of reviewers might then be to verify the accuracy and relevancy of citations rather than the worth of the article itself. The reviewers task would be easier and more objective task than the present one, and well suited as a task for graduate students and new professionals. Gordon Becker UNO Omaha NE 68182 becker@unoma1.BITNET ------------------------------ PSYCOLOQUY is sponsored by the Science Directorate of the American Psychological Association (202) 955-7653 Co-Editors: (scientific discussion) (professional/clinical discussion) Stevan Harnad Perry London, Dean, Cary Cherniss (Assoc Ed.) Psychology Department Graduate School of Applied Graduate School of Applied Princeton University and Professional Psychology and Professional Psychology Rutgers University Rutgers University Assistant Editors: Malcolm Bauer John Pizutelli Psychology Department Psychology Department Princeton University Rutgers University End of PSYCOLOQUY Digest ******************************