1


A declarative account of deletion phenomena in English phonetics and phonology.
Deletion is a process invoked in almost all phonologies. However, the growing research paradigm of 
Declarative Phonology (DP) (eg. Bird 1995, Coleman 1995) is formally more constrained than traditional 
phonologies, and rejects the use of deletion as a valid process in phonological accounts. The proposed research 
considers whether deletion is a necessary process in a DP account of spoken English, or whether apparent 
instances of deletion can be dealt with by a more sophisticated view of phonetic interpretation. It draws on 
DP, and also on the current trend in phonetic analysis towards detailed, non-segmental descriptions of 
phonetic material. The Department of Language & Linguistic Science has proven expertise in both these 
fields (Kelly & Local 1989; Ogden 1995a, b), and has successfully produced a speech synthesis system which 
employs both a declarative phonology and non-segmental phonetic interpretation (Coleman 1992; Local 1992; 
Ogden 1992).
The specific aims and objectives of the research are:
(i)	to provide a non-destructive account of deletion phenomena in declarative phonology
(ii)	to provide thorough, detailed and novel phonetic descriptions of reduced forms of spoken British 
English;
(iii)	to provide a constraint-based account of naturally occurring speech which integrates declarative 
phonology with non-segmental phonetic descriptions.
Background
Coleman (1995a: 354) describes Declarative Phonology as a phonology which makes use of four independent 
constraints, parallel to those found in syntactic formalisms such as HPSG or GPSG: 
1.	structural constraints (eg. constraints on syllable, lexical, metrical, morpheme structure)
2.	constraints on the linear order of sister constituents
3.	constraints on the set of phonological units
4.	a set of principles of phonetic interpretation which serve as a constraint on the semantics of 
the phonological formalism.
In order to implement these constraints, two mechanisms are used: unification, and structure building. Many 
operations that are standardly treated as structure-changing can be reanalysed as structure-building.
Unification is a well understood technique which has definable logical properties and parsing mechanisms. 
Thus declarative phonological analyses are computationally testable. Furthermore, although frameworks 
such as HPSG have relatively well-developed accounts of syntactic and semantic data, they have as yet 
only sketchy phonological and phonetic components. Therefore work on phonetic and phonological material 
will help determine whether the phonological component can truly be integrated into HPSG signs.
Deletion
The constraint that deletion is not allowed is of particular relevance to phonology, since it is a commonly 
invoked structure-changing operation. Furthermore, deletion is a procedure that is commonly invoked in all 
kinds of linguistic analyses. 
Traditionally citation forms provide the basis of the underlying phonological representation of words 
(Chomsky and Halle, 1968; Linell, 1977; Lass, 1984; Giegerich 1994), including function (closed-class) words. 
To relate casual (or reduced) and citation forms of any given lexical item, it is usual to formulate rules in 
terms of potentially destructive processes such as reduction, elision, deletion. Since citation forms typically 
contain �more� phonetic material than the casual forms, the derivational rules that link the two are 
necessarily destructive. The examples given below (from Jones 1960, Brown 1977, and Manuel et al, 1992) 
illustrate:
	�citation�	�casual�	informal description of rule
	hav	�v	deletion of h; reduction of a
	fr�m	fr�m frm	reduction of �; deletion of vowel
	wIl	l	possible loss of syllabicity; deletion of wI
	s�p�t	sp�t	deletion of vowel
	INklud	Nklud	deletion of vowel
	IfDe�kUdbi	f�Tek�bI	reduction and deletion of vowels; deletion of 
consonants; loss of voicing in dental fricative
On a formal level, there is no distinction to be drawn between elision, reduction and deletion; all involve 
rules which remove information, and which are not permissible in declarative phonology. As is clear from 
the examples above, using citation forms gives rise to accounts which necessarily make extensive use of 
deletion, although this particular process is named differently in different theories (eg. �delinking� in 
autosegmental phonology). 
Deletion�despite its plausibility�raises problems at a number of levels:
(1) Linguistic-perceptual: If material is deleted, how are listeners able to reconstruct what words have been 
said? How can listeners unravel the signal and know what the �underlying� citation forms might be, given an 
input which contains less phonetic information than the citation form? Klatt (1989) argues that deletions are 
a serious problem for any theory of speech perception and speech recognition.
(2) Acoustic phonetic: Deletion should result in a phonetic signal lacking in large amounts of useful 
information. However, recent research (eg. Kelly & Local 1989; Manuel 1992; Simpson 1992; Hawkins & 
Slater 1994; Manuel 1995) shows that this assumption is questionable: there is phonetic detail present in 
speech which is not readily apparent in any cursory examination which imposes a small number of categories 
on to the material, thus suggesting phonetic simplicity which may be illusory.
(3) Formal-computational: Rule formalisms that allow deletion may lead to undecidable grammars. 
Grammars with deletion are of the class of unrestricted rewrite systems, and are thus completely 
unconstrained. �If all we can say about a grammar of a natural language is that it is an unrestricted rewriting 
system, we have said nothing of any interest�, Chomsky (1963: 360). 
Phonetic detail
The fact that phonetic interpretation in DP serves as a constraint on the semantics of the phonological 
formalism makes abstraction in the phonology unproblematic; all phonological representations must be given 
a phonetic interpretation.
In general, however, research in declarative phonology has not relied on much original phonetic observation, 
but has tended instead to recast procedural phonological analyses in declarative terms (eg. Klein 1993, 
Scobbie 1993). The output of the phonology is still commonly represented as a string of broad phonetic 
symbols. The result of this is that although the phonological formalism is more constrained, the nature of 
the phonic data is generally not questioned, and the phonetic substance remains largely unaccounted for (but 
see Coleman 1995b, where original phonetic observations are made on Berber). According to Coleman�s 
criteria, then, one of the main constraints of DP (number 4  above) is not currently being properly exploited. 
The proposed research will integrate phonetic observation with a declarative phonological analysis.
The particular aspects of declarative phonology that this research will investigate are:
�	the claim that phonological processes can be accounted for by structure building rules only
�	the phonetic interpretation of phonological categories in British English.
Current research suggests that there are at least three ways in which cases of deletion could be re-analysed: 
(1) parametric reorganisation; (2) observation of fine phonetic detail; (3) suppletive allomorphy. 
1.	The phonological structure remains unchanged; it is the phonetic interpretation which changes 
through the temporal reorganisation of phonetic parameters. 
This proposal for dealing with deletion has already been made in the DP literature (Bird & Klein 1990; 
Coleman 1992. It relies crucially on the understanding that phonetic interpretation provides not just 
parameter values, but also values which describe the mutual temporal phasing of phonetic parameters 
(Ogden 1992, 1995b; Local & Ogden 1994); and that any phonological term can be given varying phonetic 
interpretations depending on structural context. In this way, Coleman (1992) accounts for cases of vowel 
deletion and reduction in unstressed English prefixes, claiming instead that the vocalic gesture and the 
consonantal gesture are co-extensive in time in syllables where a vowel has apparently been deleted. In the 
articulatory domain, Browman & Goldstein (1992) show that contextual variation of phonetic units can be 
accounted for by underspecification and temporal overlap. The notion of phonetic reorganisation therefore 
makes extensive use of the co-production model (eg. Fowler 1977). 
This treatment of �deletion� is relatively easily tested, since it predicts that the �missing� phonetic 
parameter is not missing, but that it appears elsewhere. 
This model of temporal interpretation has already been computationally implemented (Local & Ogden 
1994), and been shown to produce successful synthesis of fast forms for English.
2.	There is fine �phonetic detail� that has typically been ignored as not essential to phoneme 
identification� (Hawkins 1995: 20). 
According to this view, deletion appears not because there is actually material deleted, but because the 
phonetic observation and generalisation are not at a fine enough level. In other words, the apparent deletion 
is a product of the symbols which are used to make a transcription that serves as the basis for phonological 
analysis. Furthermore, these symbols may not in fact capture the phonetically most significant feature. This 
appears to  be particularly true for closed-class items. This line has been pursued by eg. Manuel et al. (1992), 
Manuel (1995), a consideration of reduced forms of /D/ in American English, Zsiga (1995), a discussion of the 
articulation of various sounds of English more generally allotted to /S/, and Simpson (1992) who considers, 
among other things, the phonetic characteristics of pronouns in Suffolk English.
According to this view, phonetic description should be made at a more refined level than has so far been the 
case. This view is entirely compatible with that put forward in (1).
3.	Forms which are phonetically very different should be accounted for by suppletion.
This view is espoused by Kaisse (1985). The relation between eg [wUd] and [d] must, according to conventional 
interpretations, involve deletion rules; furthermore these rules are not productive, but apply to only a few 
closed-class items, and even then sporadically (eg. the rule that deletes [w] in �would� does not apply to 
�was�). Kaisse�s solution is to say that forms like [wUd] and [d] are suppletive allomorphs; but then the onus 
is on the analyst to state the conditions under which each allomorph might occur; and it is unclear according 
to this view why the two allomorphs should have anything at all in common, since other suppletive 
allomorphs (eg. �go� and �went�) have no phonetic resemblance.
If none of the above should prove possible for a given piece of data, then there is only one choice left:
4.	Deletion has to be accepted as a possible phonological operation.
From the point of view of declarative phonology, this view is the most challenging one. If it turns out that 
none of the other strategies above is tenable, and deletion has to be accepted as a possible phonological 
operation, then declarative phonology would suffer a severe blow. However, the problems with deletion 
identified above remain.
Research questions
The specific questions which will be addressed include:
1.	Can apparent instances of deletion be explained by variability in phonetic interpretation?
2.	Can apparent instances of deletion be accounted for by considering hitherto ignored fine phonetic 
detail?
3.	Do apparent instances of deletion have to be accounted for by appeal to suppletive allomorphy?
4.	Is deletion a necessary process in phonological theory?
Data
Data will be taken from the MARSEC, a database of spoken English funded by the ESRC, Grant No. R000 23 
3380 (Roach et al. 1993). The accent of all the speakers recorded in the database is RP, or close to RP. 
Material in the database has been tagged at a number of different levels of linguistic description, including 
dictionary pronunciation and grammatical tag. The majority of the recordings on the database come from 
radio broadcasts, and in all the recorded material amounts to six hours� worth. 
MARSEC has several advantages:
�	the quality of the recordings is high
�	it provides natural material
�	it contains material from a wide range of speakers 
�	it contains material in a wide range of speech styles
As a result, we  will be able to consider whether deletion processes occur at different rates and to differing 
degrees with different speakers. Both auditory and acoustic analysis will be possible. The speech can be said 
to be representative of a cross-section of British English speakers, rather than an idealisation.
Analysis methods
1.	The first task will be to find suitable instances of deletion phenomena. There are essentially two 
important deletion sites which we shall consider: (a) grammatical contexts (such as articles, 
pronoun+auxiliary) (b) metrically weak positions in polysyllabic lexical items. 
Preliminary work carried out shows that this work can be done relatively quickly and efficiently by using 
UNIX search mechanisms on the text files that accompany the audio files on MARSEC.
2.	The main work will be that of analysis. This falls into three separate areas.
(a) Detailed auditory phonetic observation and transcription using the IPA�
(b)  �accompanied by quantitative acoustic corroboration. 
In particular, parametric phonetic observations will be made, with particular attention to parameters such 
as formant frequency, voicing, nasalisation, voice quality, fundamental frequency, and the relative phasing 
of these and other parameters cf. Kelly & Local (1989). MARSEC files will be read into the XWAVES 
package, allowing a wide range of analysis techniques to be used.
(c) Phonological analysis of the phonetic observations.
The main part of the analysis work will be to consider how a phonological analysis can be made of the 
phonetic observations that have been made. The phonological analysis will follow the guidelines suggested 
in Kelly & Local (1989: 99ff):
�	parsimony ie. minimising the number of phonological primitives) 
�	system-symmetry  ie. looking for and eliciting symmetry in phonological systems which are 
established.
�	a strict demarcation between phonetics and phonology.
�	viewing the phonetics-phonology relation as non-derivational but declarative.
�	parametric interpretation of phonetic material
�	variable-domain interpretation of phonetic features over all kinds of domains, and not just �segment�-
sized pieces
�	accepting that phonetic parameters are of variable relevance depending on the place in the 
statement
�	congruent level interpretation: relating the phonetic detail to various levels of linguistic analysis, in 
this case to syntactic, morphological or metrical levels
�	polysystemic interpretation of material � ie. in terms of several systems rather than one overall 
system.
Expected outputs
At the phonetic level, the investigation will provide researchers with phonetically detailed and rich 
descriptions of spontaneously produced English in traditional terms of auditory (impressionistic) phonetic 
classification as well as in acoustic terms. Thus knowledge of spoken British English will be expanded.
The phonetic  observations will be in at such a level of detail as to allow for the implementation of results in 
speech synthesis. Naturalness is one of the enduring problems in the computational synthesis of speech, and 
the improvement of synthetic speech would increase the quality of life for people who use speech 
synthesisers.
At the phonological level, the research will deepen understanding of declarative and non-segmental, non-
linear phonology. The research will also shed new light on the relation between phonetics�particularly the 
phonetics of spontaneous speech�and phonology. 
It is also possible that the research into the phonological description of the pronominal and auxiliary system 
may prompt a reassessment of the syntax-phonology interface. 
The research could profitably inform those areas of linguistics which also have to deal with the zero 
category, such as research which seeks to establish whether the absence of a given feature indicates 
deletion, or some other mechanism in the grammar, eg. variationist analyses of competing grammatical 
systems, in which the different weighting of phonological vs. syntactic constraints on variable features (eg. 
/s/ ~ /�/ in plural marking) is taken to be an indicator of the underlying grammar. 
There are a number of journals suitable for disseminating this kind of work, such as Phonology, Journal of 
Phonetics, Phonetica, Speech and Language. Findings of the research will also be reported progressively 
through York Papers in Linguistics and the research papers of the Department of Language & Linguistic 
Science at the University of York. The research findings will also be disseminated at conferences.

References:
Bird, Steven (1995): Computational Phonology: A Constraint-Based Approach, Studies in Natural Language 
Processing. Cambridge: Cambridge University Press
Bird, Steven and Ewan Klein (1990). Phonological events. Journal of Linguistics, 26, 33-56. 
Browman & Goldstein (1992): �Targetless� schwa: an articulatory analysis. In G J Docherty & D R Ladd 
(eds): Papers in laboratory Phonology II: Gesture, Segment, Prosody.Cambridge: CUP.
Brown, Gillian (1977): Listening to English. Harlow: Longman�s.
Chomsky, N (1963): Formal Properties of Grammars. In R. D. Luce, R. R. Bush and E. Galanter (eds.): 
Handbook of Mathematical Psychology 2, New York: John Wiley, 323-418.
Chomsky, N & M. Halle (1968): The sound Pattern of English. New York: Harper & Row.
Coleman J. S. (1992): The phonetic interpretation of headed phonological structures containing overlapping 
constituents. Phonology 9 1-44.
Coleman, John (1995a): Declarative lexical phonology. In J. Durand and F. Katamba (eds.): Frontiers of 
Phonology. London: Longman�s. 333-382.
Coleman, John (1995b): Declarative syllabification in Tashlhit Berber. In J. Durand & B. Laks (eds.) Current  
Trends in Phonology: Models and Methods. CNRS: Paris X and University of Salford. Salford: 
University of Salford Publications.
Fowler, C. A. (1977): Timing control in speech production.  Bloomington: IULC.
Giegerich, Heinz J. (1994): English Phonology. Cambridge: CUP.
Hawkins, S & A. Slater (1994): Spread of CV and V-to-V coarticulation in British English: Implications for 
the intelligibity of synthetic speech. Proc. ICSLP 1994, 1, 57-60.
Hawkins, Sarah (1995): Arguments for a non-segmental view of speech perception. Proceedings of the XIIIth 
International Congress of Phonetic Sciences, 3, 18-25.
Jones, Daniel (1960): An outline of English phonetics (ninth edition). Cambridge: Heffer�s.
Kaisse, E. M. (1985): Connexted Speech. Orlando: Academic Press.
Kelly, John & John Local (1989): Doing Phonology. Manchester: University Press. Klatt (1989)
Klatt, D. H. (1989): Review of selected models of speech perception. In W. Marslen-Wilson (ed.) Lexical  
Representation and process. Cambdige: MIT Press 169-226.
Klein, Ewan (1993): Sierra Miwok verb stems. Edinburgh Working Papers in Cognitive Science 8: Cognitive 
Phonology, 19-35.
Lass, Roger (1984): Phonology: An introduction to basic concepts. Cambridge: CUP.
Linell, Per (1979): Psychological reality in Phonology. Cambridge: Cambridge University Press.Lass, 1984
Local, John & Richard Ogden (1994): A model of timing for non-segmental phonological structure. Proceedings 
of ESCA/IEEE workshop on Speech Synthesis. To appear in a book of selected proceedings.
Manuel, Sharon Y, Stefanie Shattuck-Hufnagel, Marie Huffman, Kenneth N. Stevens, Rolf Carlsson & Sheri 
Hunnicutt (1992): Studies of vowel and consonant reduction. Proceedings of the Internaitonal 
Conference on Speech and Language Processing. 2, 943-946.
Manuel, S. Y. (1995): Speakers nasalise /D/ after /n/, but still hear /D/. Journal of Phonetics 23, 453-476.
Ogden, Richard A (1992): Parametric Interpretation in YorkTalk. York Papers in Linguistics 16, 81-99.
Ogden, Richard (1995a): �Where� is timing? A response to Caroline Smith.Papers in Laboratory Phonology 4, 
Cambridge: CUP, 223-234.
Ogden, Richard (1995b): Palatality as a Prosody in Tundra Nenets. Proceedings of the XIIIth International 
Congress of Phonetic Sciences, 1, 54-57.
Roach, Peter,  Gerry Knowles, TamasVaradi, Simon Arnfield (1993): MARSEC: A machine-readable Spoken 
English Corpus. Journal of the International Phonetics Association 23, 47-53.
Scobbie, James M. (1993): Issues in constraint violation and conflict. Edinburgh Working Papers in Cognitive 
Science 8: Cognitive Phonology, 37-54.Simpson (1992)
Zsiga, Elizabeth C. (1995): An acoustic and electropalatographic study of lexical and post-lexical 
palatalization in American English. In Amalia Arvaniti & Bruce Connell (eds) Papers in Laboratory 
Phonology 4. Cambridge: CUP, 282-302.