1: \documentclass[12pt]{article}
2: \usepackage{times,apalike,url}
3: \title{Computational Phonology}
4: \author{Steven Bird}
5: \date{University of Pennsylvania}
6: \bibliographystyle{apalike}
7:
8: \begin{document}
9: \maketitle
10:
11: Phonology, as it is practiced, is deeply computational.
12: Phonological analysis is data-intensive and the resulting
13: models are nothing other than specialized data structures and algorithms.
14: In the past, phonological computation -- managing data and developing
15: analyses -- was done manually with pencil and paper. Increasingly,
16: with the proliferation of affordable computers, IPA fonts and drawing software,
17: phonologists are seeking to move their computation work online.
18: {\em Computational Phonology} provides the theoretical and technological
19: framework for this migration, building on methodologies and tools from
20: computational linguistics. This piece consists of an {\em apology} for
21: computational phonology, a history, and an overview of current research.
22:
23: {\bf Documentation and Description.}
24: Phonological data is of essentially three types:
25: texts, wordlists and paradigms.
26: A text is any phonetically transcribed narrative or conversation.
27: A wordlist is any compilation of linguistic forms which can be uttered
28: in isolation, with information about pronunciation and meaning.
29: A paradigm is broadly construed to mean any
30: tabulation of words or phrases which illustrates
31: contrasts and systematic variation.
32: Any of these data types may be {\em annotated} with more abstract
33: information originating from a phonological theory, such
34: as syllable boundaries, stress marks and prosodic structure.
35: Additionally, any of these data types may be associated with
36: recordings of audio, video or physiological signals.
37: Digitizing this documentation and description brings all the
38: different media types together, makes the cross-links
39: navigable, and opens up many new possibilities for
40: management, access and preservation.
41:
42: {\bf Exploration and Analysis.}
43: The data types described above are closely interconnected in
44: phonological practice.
45: For instance, the discovery of a new word in a text
46: may require an update to the lexicon and
47: the construction of a new paradigm (e.g. to correctly
48: classify the word). Fresh insights may lead to new annotations
49: and further elicitation, closing the loop in this perpetual,
50: exploratory process.
51: Phonological analysis typically involves defining a formal model,
52: systematically testing it against data, and comparing it with other models.
53: (In some cases, the model may be incorporated into a software system, e.g. for
54: generating natural intonation in a text-to-speech system.)
55: In this exploration and analysis --
56: sorting, searching, tabulating, defining, testing and comparing --
57: the principal task is computational.
58:
59: Perhaps the earliest work in computational phonology was Bobrow and
60: Fraser's {\em Phonological Rule Tester} \cite{BobrowFraser68}, an
61: implementation of SPE designed to ``alleviate the problem of rule
62: evaluation.'' Shortly afterwards Johnson showed that, while SPE rules
63: resemble general rewriting systems at the top of the
64: Chomsky hierarchy, the way SPE rules are used in practice only requires
65: finite state power \cite{Johnson72}.
66: Independently, Kaplan and Kay discovered the connections between SPE
67: grammars and finite state transducers in the 70's and 80's, and laid down
68: a complete algebraic foundation (ultimately reported in
69: \cite{KaplanKay94}). Significant implementations followed, including
70: \cite{Koskenniemi83b,BeesleyKarttunen02}. Attempts to apply finite
71: state devices to Autosegmental Phonology have largely foundered,
72: but applications to Optimality Theory are thriving.
73:
74: While finite-state phonology fixated on SPE, generative phonology continued
75: its rapid evolution. The discovery of rule ``conspiracies''
76: \cite{Kisseberth70} and the abstractness controversy \cite{Koutsoudas74},
77: lead to calls for the reintroduction of {\em surface structure
78: constraints}. Many theories arose from the fallout; most notable
79: for its computational ramifications was
80: Montague Phonology \cite{Wheeler81}. This model adapted new lexicalist
81: formalisms from syntax and semantics, providing a {\em declarative}
82: (as opposed to {\em procedural}) account of phonological well-formedness,
83: and providing the first computational account of underspecification (where the
84: phonological content of a lexical entry is incompletely specified,
85: to be filled in during a derivation). From these beginnings, {\em Declarative
86: Phonology} was born, and subsequent work provided a mathematical foundation
87: in first-order logic \cite{Bird95} and phonetic interpretation with links
88: to Firthian prosodic analysis and speech synthesis \cite{Coleman97},
89: with implementations generally in the Prolog programming language.
90:
91: A third major strand of development, complementing the finite state and
92: declarative models, is best characterized as statistical. It seeks to
93: apply neural networks, information theory, and weighted automata
94: in the automatic discovery of phonological information.
95: Gasser trained a recurrent neural network to recognize syllables and to repair
96: ill-formed syllables \cite{Gasser92a}.
97: Ellison showed how a technique from information theory called
98: MDL -- {\em minimum description length} -- could be applied to
99: automatically identify syllable boundaries in phonemically transcribed
100: texts \cite{Ellison92b}.
101: Many researchers apply Markov models (a kind of weighted automata) in
102: speech recognition, mapping speech recordings to phonetic transcriptions
103: and thence to orthographic words, using large, phonetically annotated
104: corpora as training data (e.g. TIMIT \cite{TIMIT86}).
105:
106: Four key areas of ongoing research in computational phonology are in
107: Optimality Theory, automatic learning, interfaces to grammar and
108: phonetics, and supporting phonological description in the field.
109: Comprehensive references to online research papers in this areas may
110: be found on the SIGPHON website.
111:
112: Computational phonology is generating sophisticated and rigorous ways for
113: creating, exploring and disseminating multidimensional phonological
114: information, encompassing primary recordings, texts, wordlists, paradigms,
115: theories and analyses. As phonologists adopt the computational methods
116: described above, extending and adapting them as needed, the consequences
117: for the discipline will be increased accessibility, accountability, and
118: stability of empirical research.
119:
120: {\bf Resources.}
121: The Association for Computational Linguistics
122: (ACL) has a special interest group in computational phonology (SIGPHON) with
123: a homepage at \url{http://www.cogsci.ed.ac.uk/sigphon/}.
124: The website contains online proceedings for SIGPHON workshops
125: and information about relevant books, dissertations and articles.
126: A special issue of {\it Computational Linguistics} devoted to
127: computational phonology was published in 1994 \cite{Bird94}.
128:
129: \raggedright
130:
131: \begin{thebibliography}{}
132:
133: \bibitem[Beesley and Karttunen, 2002]{BeesleyKarttunen02}
134: Beesley, K.~R. and Karttunen, L. (2002).
135: \newblock {\em Finite-State Morphology: Xerox Tools and Techniques}.
136: \newblock Studies in Natural Language Processing. Cambridge University Press.
137:
138: \bibitem[Bird, 1994]{Bird94}
139: Bird, S., editor (1994).
140: \newblock {\em Computational Linguistics: Special Issue on Computational
141: Phonology}, volume 20(3).
142: \newblock MIT Press.
143:
144: \bibitem[Bird, 1995]{Bird95}
145: Bird, S. (1995).
146: \newblock {\em Computational Phonology: A Constraint-Based Approach}.
147: \newblock Studies in Natural Language Processing. Cambridge University Press.
148:
149: \bibitem[Bobrow and Fraser, 1968]{BobrowFraser68}
150: Bobrow, D.~G. and Fraser, J.~B. (1968).
151: \newblock A phonological rule tester.
152: \newblock {\em Communications of the ACM}, 11:766--72.
153:
154: \bibitem[Coleman, 1997]{Coleman97}
155: Coleman, J.~S. (1997).
156: \newblock {\em Phonological Representations --- their names, forms and powers}.
157: \newblock Cambridge Studies in Linguistics. Cambridge University Press.
158:
159: \bibitem[Ellison, 1992]{Ellison92b}
160: Ellison, T.~M. (1992).
161: \newblock {\em Machine Learning of Phonological Structure}.
162: \newblock PhD thesis, University of Western Australia.
163:
164: \bibitem[Garofolo et~al., 1986]{TIMIT86}
165: Garofolo, J.~S., Lamel, L.~F., Fisher, W.~M., Fiscus, J.~G., Pallett, D.~S.,
166: and Dahlgren, N.~L. (1986).
167: \newblock {\em The {DARPA TIMIT} Acoustic-Phonetic Continuous Speech Corpus
168: {CDROM}}.
169: \newblock NIST.
170: \newblock \url{http://www.ldc.upenn.edu/Catalog/LDC93S1.html}.
171:
172: \bibitem[Gasser, 1992]{Gasser92a}
173: Gasser, M. (1992).
174: \newblock Learning distributed representations for syllables.
175: \newblock In {\em Proceedings of the Fourteenth Annual Conference of the
176: Cognitive Science Society}, pages 396--401. Hillsdale NJ: Lawrence Erlbaum
177: Associates.
178:
179: \bibitem[Johnson, 1972]{Johnson72}
180: Johnson, C.~D. (1972).
181: \newblock {\em Formal Aspects of Phonological Description}.
182: \newblock The Hague: Mouton.
183:
184: \bibitem[Kaplan and Kay, 1994]{KaplanKay94}
185: Kaplan, R.~M. and Kay, M. (1994).
186: \newblock Regular models of phonological rule systems.
187: \newblock {\em Computational Linguistics}, 20:331--78.
188:
189: \bibitem[Kisseberth, 1970]{Kisseberth70}
190: Kisseberth, C.~W. (1970).
191: \newblock On the functional unity of phonological rules.
192: \newblock {\em Linguistic Inquiry}, 1:291--306.
193:
194: \bibitem[Koskenniemi, 1983]{Koskenniemi83b}
195: Koskenniemi, K. (1983).
196: \newblock {\em Two-Level Morphology: A General Computational Model for
197: Word-Form Recognition and Production}.
198: \newblock PhD thesis, University of Helsinki.
199:
200: \bibitem[Koutsoudas et~al., 1974]{Koutsoudas74}
201: Koutsoudas, A., Sanders, G., and Noll, C. (1974).
202: \newblock The application of phonological rules.
203: \newblock {\em Language}, 50:1--28.
204:
205: \bibitem[Wheeler, 1981]{Wheeler81}
206: Wheeler, D.~W. (1981).
207: \newblock {\em Aspects of a Categorial Theory of Phonology}.
208: \newblock PhD thesis, University of Massachusetts at Amherst.
209:
210: \end{thebibliography}
211:
212:
213:
214: \end{document}
215: