LINGUIST List 19.1859
|
Wed Jun 11 2008
Review: Writing Systems: Sproat (2000)
Editor for this issue: Randall Eggert
<randy linguistlist.org>
|
This LINGUIST List issue is a review of a book published by one of our
supporting publishers, commissioned by our book review editorial staff. We
welcome discussion of this book review on the list, and particularly invite
the author(s) or editor(s) of this book to join in. To start a discussion of
this book, you can use the
Discussion form on the LINGUIST List website. For
the subject of the discussion, specify "Book Review" and the issue number of
this review. If you are interested in reviewing a book for LINGUIST, look for
the most recent posting with the subject "Reviews: AVAILABLE FOR REVIEW", and
follow the instructions at the top of the message. You can also contact the
book review staff directly.
|
Directory
1. Earl
Herrick,
Computational Theory of Writing Systems
Message 1: Computational Theory of Writing Systems
|
Date: 10-Jun-2008
From: Earl Herrick <graphonomistsbcglobal.net>
Subject: Computational Theory of Writing Systems
 E-mail this message to a friend
Announced at http://linguistlist.org/issues/18/18-149.html
AUTHOR: Sproat, Richard
TITLE: A Computational Theory of Writing Systems
SERIES: Studies in Natural Language Processing
PUBLISHER: Cambridge University Press
YEAR: 2000
Earl M. Herrick, Dept. of Language and Literature, Texas A&M University-Kingsville
We linguists must remember that we use two different definitions of the word
''grapheme''. They have been nicely distinguished by Kohrt (1986); I have
paraphrased them in one of my forthcoming works as (1) something that is
entirely inside the graphonomy of a language, which Kohrt calls the ''analogical
view'' of a grapheme, and (2) something inside the phonology of a language that
is related in a certain way to its graphonomy, which Kohrt calls the
''referential view'' of a grapheme.
Sproat uses the term ''grapheme'' with its referential meaning. (I use it with its
analogical meaning in my own work.) Doing so allows Sproat to use it for talking
about whatever bits of speech are referred to by any given mark that he is
discussing without having to analyze the mark itself--which is, after all, not
what he is interested in doing. And by a ''writing system'' he means a method of
writing down marks that refer to sounds, and does not mean the marks themselves,
or their differences that distinguish them from one another. In the same spirit,
he uses the term ''linguistic'' to refer to speech rather than to writing, as when
he asks ''what linguistic elements do written symbols encode?'' There is nothing
wrong with doing this, as long as the reader knows what the writer means.
Sproat's work has been in text-to-speech synthesis, and he is a professor both
of linguistics and of electrical & computer engineering.
SUMMARY
In chapter 1, ''Reading Devices'' Sproat says that his topic is how to computerize
text-to-speech synthesis (TTS). He assumes that the input for his procedures
exists in some electronic form and that its output is digital representation of
speech. (He gives references for those who may be interested in optical
character recognition and speech synthesis as input and output for his own
model.) He notes that to pronounce a word aloud it is often necessary to know a
lot about the language that is not present in the word's written form. And he
also notes that to do so, things may have to be accounted for that people are
often unaware of, such as the fact that many writing systems do not have spaces
between their words.
As an example, Sproat discusses a pair of Russian words that are spelled the
same but differ in their stress placement and also (consequently for Russian) in
their reduced vowels. He describes them and a Chinese word by both an
Attribute-Value Matrix (AVM) and an annotation graph, both of which show the
words' pronunciations and their orthographies. He then presents definitions and
axioms in logical notation, and he comes to the central claims of his theory,
which are that (1) the mapping between the Orthographically Relevant Level (ORL)
of a language and the written characters in its words is regular, and that, (2)
for a given writing system and a given language, the ORL represents a consistent
level of linguistic representation.
As further issues, Sproat discusses why a theory of writing systems should be
constrained and why a study of writing systems should rely on a segmental
analysis of spoken language. He then concludes the chapter with an outline of
terminology and conventions that he will use, and he adds an appendix on
finite-state automata and transducers.
In chapter 2, ''Regularity'', Sproat deals with the fact that, although spoken
utterances exist in one temporal dimension, written utterances exist in two
spatial dimensions. He refers briefly to ''ordinary (string-based) regular
languages'', which can be written with only a sequence catenator because the
sequence of written characters matches the one-dimensionality of speaking. (He
does refer to the problem of distinguishing between apostrophes and commas by
mentioning their heights, but he merely mentions the tone diacritics of
languages such as Thai, Vietnamese, and Navajo in a footnote, p. 4.) Sproat then
discusses ''planar regular languages'', in which the written characters must have
their relationships described in two dimensions and therefore require a richer
set of catenation operators. He introduces the notion of the Small Linguistic
Unit (SLU), within which the sequence of characters does not have to conform to
their ''macroscopic'' (line- and document-level) order, and he provides a
systematic notation for expressing catenation, both of characters within an SLU
and of SLUs macroscopically.
Sproat then gives examples of four writing systems for which the catenation of
components within the SLUs needs to be stated: Korean Hankul, Devanagari, Pahawh
Hmong, and Chinese. The Hankul characters for (historical) syllables are
arranged within square spaces, and those characters are arranged there by a
simple rule, according to whether they are horizontal or vertical in shape. In
Devanagari, initial vowels are written by separate characters. For every other
graphic syllable (defined as everything after one phonological vowel and
including the next such vowel, and not necessarily forming a phonological
syllable) the consonants of the syllable, if there are more than one, are
combined into a ligature, and the vowel, if not schwa, is written by a character
in a certain place in the orbit around the consonant character or ligature. No
vowel character is written when the spoken vowel is schwa. For the Pahawh Hmong
system (Smalley et al., 1990), the characters are divided into two groups:
onset-consonant and vowel-cum-lexical-tone. For each written syllable, the
vowel-cum-tone character is written before the onset character, although their
corresponding sounds are pronounced in the opposite order. For Chinese, Sproat
agrees that its complex characters can be analyzed into many levels, and he
introduces additional notation for their internal structure that is based on the
catenation operators he has already used. He notes that almost every Chinese
character can be divided into a phonetic component, giving some information
about its pronunciation, and a semantic component, giving clues to its meaning,
and he also notes that each character has a ''determining component'' (its
semantic one, unless its phonetic one belongs to a certain set of eight) that
controls the placement of its other components. He also notes that the more
regularly Chinese pronounced characters are more regular in their structure.
Sproat mentions possible counterexamples from Ancient Egyptian, Spanish, and Mayan.
Finally, Sproat discusses ''macroscopic catenation: text direction''. He assumes
that an ordinary written text can be modeled by a ''virtual tape'' which is
arranged on a page in rows or in columns, and which, when it reaches the edge of
a page, is cut and continued next to the previous row or column. In English,
this ''tape'' runs from left to right, in Hebrew it runs from right to left, and
in Chinese it runs from top to bottom. Sproat also mentions boustrophedon
writing and points out that shop signs may show variations on the basic way that
a text is arranged in a language.
In chapter 3, ''ORL [Orthographically Relevant Level] Depth and Consistency'',
Sproat considers the levels within the phonology that are represented by various
writing systems and the consistency with which those levels are represented, and
he mentions that the spelling of some words must be ''lexically marked'', i.e.
specified without reliance on the phonology.
He begins with a case study of Russian and Belarusian, which he says form a near
minimal pair for this comparison, each showing great internal consistency, but
with ORLs that have different depth.
For English, Sproat acknowledges the existence of Chomsky & Halle 1968
(hereafter SPE), but he says that much of it merely shows ''personal taste about
how writing systems should be designed'', and he mentions that Sampson 1985 has
pointed out serious defects in SPE. Sproat remarks that ''the system of English
spelling is a great deal more chaotic than that of [...] almost any other
language that uses a script whose original design was purely phonological.''
However, he rejects the often-expressed idea that English has a logographic
writing system because the evidence for it is so inconsistent, unlike the
consistent logographic elements in the Chinese writing system. Sproat gives a
32-page appendix of words of the sort that are central for the arguments
presented in SPE, giving for each of them a ''Deep ORL'' and a ''Shallow ORL'', and
he mentions that the addition of a great number of other English words would
probably make the argument for a deep ORL less convincing, except for the
question of how to write reduced vowels.
Sproat also discusses the devoicing of dental obstruents in certain environments
in Serbo-Croatian, and he presents experimental evidence that casts doubt on the
standard treatment of this phenomenon, although saying that it needs further
investigation. He discusses a possible example of cyclicity in Dutch, and
concludes that his theory has no problem with it if the cyclicity is internal to
the orthography. Finally, he discusses surface orthographic constraints, and
notes that they can be handled by environmental rules or lexical marking within
the orthography.
In chapter 4, ''Linguistic Elements'', Sproat asks about the range of linguistic
(phonological) elements that can be represented by writing systems. He looks at
the influential taxonomies of writing systems presented by Gelb (1963), Sampson
(1985), and DeFrancis (1989). He dismisses Gelb as teleological and outmoded, he
presents Sampson's and DeFrancis's tree-shaped taxonomies, and he lists
DeFrancis's disagreements with Sampson. He then presents his own taxonomy of
writing systems, for which he uses two dimensions, with the parameters ''amount
of logography'' and ''type of phonography''.
In considering Chinese, Sproat concentrates on the semantic-phonetic compounds
that are the vast majority of Chinese characters, and notes that for such
characters ''the phonological information provided by the phonetic component is
sometimes perfect (only a few), frequently only partial (by far the greatest
number) , and in some cases completely useless (only a few)''. He therefore says
that ''it is much more useful to view [Chinese writing] as an imperfect
phonographic system with additional logographic attributes, than it is to view
it as a wholly logographic system''. And he shows how logographic elements are
used when writing disyllabic Chinese morphemes.
For the Japanese writing system, Sproat describes the complications that arise
from its many layers of borrowing from the Chinese writing system, resulting in
characters most of which are logographic, because a Japanese reader must simply
memorize the association between the sounds and the marks. But he also notes
that the use of kanji (Chinese characters) has declined steadily during the
twentieth century, as more and more people have become literate. Finally, Sproat
mentions written characters in some languages that show plurality of meaning,
reduplication of sounds, and the zero pronunciation of other characters.
In chapter 5, ''Psycholinguistic Evidence'', Sproat asks what support there is in
the psycholinguistic literature for the ''psychological reality'' of the model he
proposes. He notes that there is little consistency in that literature, and he
does not suggest that the computational devices he proposes actually exist in
people's heads. He asks rather how the macroscopic properties of his model
compare with what that literature has found, especially with respect to two
questions: (1) whether the relationship between orthography and ''linguistic
form'' is the same for all writing systems, and (2) whether the ''Orthographical
Depth Hypothesis'' (ODH)--which claims that languages with ''deep'' orthographies
such as English require readers to read by going through the lexicon while
languages with ''shallow'' orthographies such as Serbo-Croatian allow readers to
go directly from the graphonomy to the phonology--is valid. (Sproat also notes
that, although Serbo-Croatian is often adduced as such a language, it does not
write lexical stress, and that Spanish would be a better example.)
Sproat claims that multiple routes from the written marks to the phonology exist
for all written languages. He cites arguments for and against the ODH and finds
that for both Chinese and Japanese there is evidence that readers use such
routes. (He notes as a possibly more familiar example that literates in English
know how to pronounce the letter string without having to think of the
word they are pronouncing.)
Sproat then considers the ''connectionist'' models that assume large numbers of
simple, but massively interconnected, units. He mentions Seidenberg and
McClelland (1989) as a classic statement of that approach, he summarizes their
model, and he also mentions more recent work. He points out a defect in their
model, he says that there is little reason for it to supersede other models of
reading, and he concludes that the overall architecture of his own model is at
least not at odds with what we know about human orthographic processing.
In chapter 6, ''Further Issues'', Sproat discusses some complications that can
arise when trying to turn a written text into an internal linguistic
representation, and he mentions Manx Gaelic as an example of an orthography
constructed so as to be similar to that of another language (in this case,
English). He discusses how the 1995 reforms in Dutch have introduced
morphological and semantic complications into its spelling. He discusses the
complicated relationships that exist in many languages between numerical
notations and spoken number names. He discusses the problems involved in
pronouncing ''abbreviatory devices'' (a term that he uses because ''abbreviation'',
''acronym'', and such terms have been used with so many different meanings). And
he discusses how various languages pronounce logograms such as , <$> and
<%> and the fact that some terms such as ''NATO'' have pronunciations based on
their abbreviations.
Sproat also mentions that many linguists such as Vachek of the Prague school and
some British linguists (although he does not mention Halliday 1989) have wanted
to treat written communication separately from and in parallel to spoken
communication. Sproat again emphasizes the fact that written texts are arranged
in two dimensions, while spoken texts are arranged in only one dimension, and he
asks whether mathematical notation should be regarded as language. He concludes
that it is a matter of definition whether written texts are to be considered as
linguistic, and he points out that he is dealing in the present book only with
mapping from written to spoken forms.
Finally, Sproat points out that all previous work has only touched on what he
discusses, and that he hopes other researchers will carry on the study of this
topic. He mentions that SPE was not underpinned by any theory of orthography;
similarly, he mentions that a large number of workers in speech technology do
not realize the grammatical and semantic elements that they must consider. And
he repeats that, for most languages, ''letter strings [...] do encode
pronunciation, but only in combination with other information that cannot be
computed from the letter string alone''.
EVALUATION
Sproat's monograph is the first to formally and systematically explore one
direction of the relationship between spoken and written language. It surely
must be taken as the basis for any such work in the future. The notation that he
provides is powerful, he goes thoroughly into certain aspects of the
relationship that he models, and he mentions the places where he sees that more
work must be done.
As befits the first monograph in a new field, Sproat gives copious references to
those who have worked in fields related to his and to the sources which he has
drawn on for data. He borrows occasionally from various versions of the
Chomskyan tradition, he creates new notations for some of the relationships he
discusses, and he frequently makes logical statements in algebraic notation;
however, all of his terminology is readily intelligible to those who will
acquaint themselves with it. As Gleason (1976) has pointed out, we have a
professional metalanguage which is composed of bits and pieces from various
theories but which we all recognize and use. Sproat has provided us some more
very useful items for our metalanguage.
In the best of all possible worlds, for every language that uses more than one
communication channel, we would describe on its own terms each communication
channel that it uses, whether spoken, written, or signed, and we would also
describe all of the relationships that exist in both directions between all of
the communication channels that it uses. Unfortunately, we are not there yet.
Spoken sounds have been well studied on their own terms, but the study of
written marks and gestured signs on their own terms, and the relationships among
all of these communication channels, is just beginning.
REFERENCES
Chomsky, Noam and Morris Halle. 1968. _The sound pattern of English_. New York:
Harper and Row.
DeFrancis, John. 1989. _Visible speech: the diverse oneness of writing systems_.
Honolulu: Univ. of Hawaii Press.
Gelb, I. J. 1963. _A study of writing_. Chicago: Univ. of Chicago Press.
Gleason, H. A., Jr. 1976. ''Continuity in linguistics.'' _LACUS Forum_ 2:3-16.
Halliday, M. A. K. 1989. _Spoken and written language_. 2nd ed. Oxford: Oxford
Univ. Press.
Kohrt, Manfred. 1986. ''The term 'grapheme' in the history and theory of
linguistics.'' In Gerhart Augst, ed. _New trends in graphemics and orthography_.
Berlin: DeGruyter, 80-96.
Sampson, Geoffrey. 1985. _Writing systems_. Stanford CA: Stanford Univ. Press.
Seidenberg, Mark, and James McClelland. 1989. ''A distributed, developmental
model of visual word recognition and naming.'' Psychological Review 96:523-568.
Smalley, William, Chia Koua Vang, and Gnia Yee Yang. 1990. _Mother of writing:
the origin and development of a Hmong Messianic Script_. Chicago: Univ. of
Chicago Press.
ABOUT THE REVIEWER
Earl M. Herrick has a Ph.D. in linguistics and is emeritus professor at Texas
A&M University-Kingsville. His 1966 M.A. thesis, ''A linguistic description of
Roman alphabets'' (Hartford Studies in Linguistics 19), which he wrote under H.
A. Gleason, Jr., remains the only analysis ever published of the features that
distinguish the characters of those alphabets from one another, although its
stratificational notation is now antiquated. He has since published a
considerable number of papers on graphonomy and on stratificational theory in
LACUS Forum and in Visible Language, and he has promised a manuscript on the
graphonomy of English to Springer.
Read more issues|LINGUIST home page|Top of issue
|
|
Please report any bad links or misclassified data
LINGUIST Homepage | Read
LINGUIST | Contact us
While the LINGUIST List makes every effort to ensure the linguistic relevance of sites listed
on its pages, it cannot vouch for their contents.
|
|
