You are on page 1of 17


Phonological systems
754 The bulk 01' the material here is based on Crothers (1978: Appendix m), except for Yiddish, RP, Gennan, Scots, Kabardian, and Swiss and Alsatian German (after Keller Ig61), and the Hungarian sysiern, which is courtesy of Veronika KniClsa. I have altered Crothers' noration lo conform more closely lO I PA conventions, and have conflatccI his 11'.El, 1;)01, where the capitals stand for 'mean mid' values, roughly beiween half-close and half-open, as lE;)/. One further rernark on the consequences ol"omitting long vowels Irorn Crothers' typological index is in order: he gives English (RP) as a six-V system with no interior vowels 011 the basis of 1I E C 1\ o 01 (with 11\1 interpreted as open central); but it has FIVE long vowels, one 01" thern (his 1;):1, my 13:1) interior. Thcrefore R P (even if 11\1is allowed as peripheral) should not be grouped with Persian. 7.6 Much ofthe material here is based on the systern-invenmries in **Nartey (1979) which is the rnost complete survey 01'obstruent systems available (based on a sample of over 300 languages). Germanic and Dravidian systerns frorn my own notes, Caucasian from Catford (1977b), Sindhi, Zulu from Ladefogecl (1971), Yidiji from Dixon (1977). Nartey's interest is in implicational universals, and he has little to say about symrnetry: Ior a good treatment of this, Hockett (1955). Nartey also omits glotlal fricatives (on dubious grounds: he inclucles I?I under stops); so data on Ih fil is panly from Hockeu , partly Irorn my own notes. Burmese and Amharic fricatives after Ladefoged (1971). 7(j4 Data on nasals lrom Narrey (1979), (Rischel 1974), Yiddish and Kannada. except for West Greenlanclic

Phonological processes

7(j5 Liquid systems not from Maddieson are Hawaiian, Chipewyan, Georgian (Hockett 1955), Malayalarn (Ladefoged 1971), German, English, Italian, Spanish, Kannada (my notes). 7(j(j Navaho, Hawaiian from Hockett (1971); all others my own observation. (1955), Margi frorn Ladefoged

77 Kabardian alter Catford (1977b). There have in fact been attempts 10 reduce the inventory still urther: Kuipers (1960) gives it NO vowels, bu! onlya 'feature of openness', taken as a kind ofsecondary articulation of consonants. This is pretty well demolished by Halle (1970). See discussion in Catford. 7.8 On polysystemic theory scc ** Firih ( 1948), and the ela borate discussion ofThai in Henderson (1951).

The concept ofprocess: terntinology, theory, problents Terms like 'velarization', 'palatalization', ete. are often used ambiguously and rather misleadingly. With years ofpraetiee one gets used to this, but it can be troublesome at the beginning. So can the proliferation of terms referring to proeess types, whieh are more or less traditional, but hard to trace to a definitive souree. This ehapter is a rough guide through the labyrinth of eoneepts and terminology, and an introduetion to the nature and forrnalization of some eharaeteristie inter-segment relations. First, 'palatalization' and the like. These are used in two basie ways: statically, as names for secondary artieulations (a palatalized eonsonant has superimposed [i]-eolour), and dynantically, as names for proeesses (to palatalize is to impose sueh eolour). But this isn 't as troublesome as the ambiguous use of the proeess sense: 'to palatalize' can not only be to impose [i]-eolouring, but to turn a non-palatal into a palatal. Thus not only [t] ~ [tj], but [t] ~ [e] or [k] ~ [e] are palatalizations, as traditionally are [k] ~ [tJl, [s] ~ [J], ete. The same holds for the other '<ization' terms. There's yet a Iurther ambiguity: terms like 'palatalization' can be (and often are) used to refer not only to phenomena that are clearly proeesses in a pretheoretieal sense (see below), but also to ones open lO reasonable non-proeess interpretations. A classie instanee is the eommon interpretation of segmental DlSTRIBUTIONS in terms of realizations that can be given proeess interpretations. Thus the English alternation offront and baek velars depending on voealie environment is often deseribed as 'palatalization before front vowels'. Such an interpretation is highly theory-dependent; it assumes, among other things, a 'base-and-derivation' theory of the phoneme (7.7), i.e. a belief in 'real' synehronie proeesses, the ehange of entities into others



Phonological processes
in the course of a derivation. If you believe this, then it's legitimate to talk of a process as 'what really lies behind' complementary distributions, say; otherwise it's simply a descriptive metaphor. Under any interpretation it's a fact that there are front and back velars; but whether this involves 'palatalization' in a process sense is a matter of theoretical faith and/or analytical style. The only extra-theoretical 'fact' is the nature and distribution of the segments. The only case where process terms can be used in a relatively theoryneutral sense is where the relation between input and output is TEMPORAL: i.e. in describing historical change. Take the now familiar voice-alternation in German obstruents; it happens to be a fact about the history of German that a sound change occurred that devoiced final obstruents, so that what is now [ra:t]/[ radas] was once (roughly) [ra:d]/[ra:dJs]. The change [d] -+ [t] was a real event in historical time, and it left traces of itself as a paradigmatic irregularity. In our description of current language states we tend to resolve the 'surface' irregularities by referring them to 'deeper' regularities (4.5); and thus we invoke processes that often mirror the historical changes that produced the irregularities in the first place. (Whether this is legitimate or not is debatable: see 9.3fT.) The problem of the 'genuineness' of processes (or phenomena statable as processes) becomes a bit more subtle when it comes to relations holding between speech ternpi. Thus 1 have a sequence of increasingly casual forms of SENTENCE, [sEn?tins], [sEn?tJ;ls], [se?tJ;ls] and [se?J;ls]. Does this mean that in going from one style to another '[E] turns into [e] and [n] deletes', or '(?t] becomes [?]'? Or is itjust that one particular form is appropriate for/used in one tempo rather than another? (Compare this with the Scots speaker who says thrapple in his most locally-rnarked register and throat in his more 'standard' variety. Do we want to say 'thrapple turns into throat ... '? Or that 'the speaker uses thrapple ... OR throat '?) That is: one possible non-historical interpretation of'process' is 'substitution relation'. This said, ir's perfectly possible to use process-notations and formalizations without any commitment to 'reality', as they are descriptively useful- so long as we know what we're doing, and keep the theoretical problems in mind. With these caveats as a background, I will draw on historical, distributional, and fasi-speech phenomena more or less indifTerently for illustration, following the (perhaps ultimately harmless) sloppiness ofthe scholarly tradition. 170

8.2 Assimilation and dissimilation Asshnilation and dissimilation These are catch-all terms: almost any process can count as one or the other, depending on contexto Broadly, in assimilation one segment becomes more like (or identical to) another (or two become more like each other); dissimilation is the converse. Thus if[k] -+ [x] coruext-free, this is simply spirantization; but if the same thing happens between vowels, this can count as assimilation: the stop takes on the opener stricture of its surroundings. Conversely if [r] -+ [1] context-free, this is just 'Iateralization' (the term is non-traditional); but ifit happens in a form containing another [r]' this is dissimilation (e.g. L arbor -+ Sp arbol). This said, assimilation is so common and important that various types are worth being discussed as such. 8.2.


Direction and contiguiry

The standard assimilation taxonomy involves direction; the assimilating inftuence may work either to the right or the left. Consider these English tempo variants:
(8.1) Tempo 1 (a) 1 ;;IOp~ln 2 sevan (b) 1 arm khAml1) 2 arm not Tempo oopm

2 'open' 'seven' 'I'm coming' 'I'm not'

am khAml1) am not

In (a) the inftuence moves from left to right, or forward; in (b) from right to left, or backward. This can be seen more clearIy if we reformulate:





Direction Direction


1 m~


Case (a) is progressive or perseverative asshnilation; (b) is regressive or anticipatory. Assimilations may be further categorized according to whether the segments in volved are in contact or separated by others. In (8. 1) we have contact assimilation, but there is also distant asshnilation, in which, either progressively or regressively, the inftuence moves across some intervening segment(s). The most characteristic distant assimilation is metaphony: noncontact vowel assimilation. Traditionally there are two types: 171

Phonological processes
(regressive) umlaur and (progressive) vowel Iiar-rnony (rhough some writers use 'vowel harmony') Ior both. Umlaul can be illustrated by the Germanic z-umtaut, in which (in general) back vowels frorued before a following jij or jjj, normally with one or more consonants intervening:


8.2 Assimilation and dissimilation Disregard for the moment the question of the underlying forms of the suffixes; we will return to this problem in vowel harmony, and other aspects, in 9.3, 10.2.3 The Hungarian data, though grossly oversimple (a lot of exceptions have been omitted) at least gives the general idea. Finally, there are 'bi-directional' or fusional assimilations, in which a sequence SSj (where S = 'segment') -. Sk (where k = some combination of features from i, j). A familiar example is English alveolarjpalatal sandhi:

0(:) [U(:)]

[ 0(:) Y(:)]


e ~


Stating it in features, the nature of the assimilation is clear, with the SD being [ + back] ... [ - back] and the SC [ + back] -. back] . The commonest meiaphonies are assimilatory; but there are dissimilatory ones as well. One striking case is a sequence of umlaut and 'anti-umlaur' in the ancestors of modern Tamil and Kannada, where the following rule-sequence has been reconslructed:


hrt ju: khld ju:

mrs ju:


khld3u: mi]u:

'hit you' 'kid you' 'miss you'



1 [~]

- [:]



output of a fusion is usually a 'compromise' segment: in (8.7) the alveolarity of the first element and the palatality of the second 'rneet halfway' in a retracted alveolar with a raised tongue-body (which is what a 'palato-alveolar' really is). A similar if less obvious type of compromise assimilation occurs in Sanskrit vowel sandhi (4 7, (4.18)), where /-0 + i-/-> [-e-], /-0 + u-/- [-0-], i.e.
The (8.8)

Stagen [:]


Curiously, the phone sets [i u] and [e o] are in complementary distribution al both stages in the sarne environrnent; only by assimilauon to lowness ar stage 1 and dissimilation al siage 11. Vowel harmony as a systernatic process can be illustrated from Hungarian. Here, most sufTixes have two or more allomorphs, which are conditioned by the vowel(s) of the preceding root-rnorpheme. In the simplest case, the sufTix has two allomorphs, one wirh a front and one with a back (or non-fronr) vowel, controlled by the stern:
(85) 'house'





The second element in the sequence controls backness, the first height; but height is assimilated by the second element moving 'one step' toward the height-controller joj, the result being a single segment combining the backness of one element and the (relative) height ofthe other.


'frorn inside N' hii:z-uo:1


harz-bnn ksrt-hsn

'at N'
harz-nar! ksrt-nerl

Basic assimiLation and dissimilation types



There is also a harmony involving frontnessjroundness, allomorph suffix:

(8.6) 'house' 'garden' 'squash ' 172

as in this three-

ha.z kert

'up tu N' harz-hoz




There is probably no segmental property that can't be the target of an assimilation or a dissimilation. It may be helpful to look at some major types in terms of the parameters they can be seen as responding to. (i) Place. Examples have aIread y been given of simple cases for consonants (8.1), as well as more complex fusiona] ones (8.7). If as suggested (6.3) we extend the term 'place' to cover vowel height and backness, all examples so far except roundness harmony in

Phonological processes Hungarian are place assimilations (and this goes for dissimilations as well). Diphthongization can also be assimilatory; in pre-Old English [u] was inserted between a front vowel and certain back consonants ('breaking' in the handbooks): e.g. jSCxj '1 saw' -+ [SCUX], jselxj 'seal' -+ [seulx]. This can be seen as 'protection' of a front vowel from a back environment; hence the 'transition' vowel [u] is an assirnilatory response. Later on, these diphthongs underwent an internal heightassimilation: [CU] -+ [Ca] , [eu] -+ reo]. Under a bi-segmental interpretation (5.3. 10), diphthongization of long vowels and monophthongization of diphthongs are relevant. If [i:] = [ii] and [u:] = [uu], then the diphthongization ofME ji: u:j in the Great Vowel Shift (7.2) is dissimilation: jiij-+ jeij, juuj-+ jouj. By the same token, the monophthongization ofME jauj to j:J:j in Early Modern English (e.g. law jb:j from jlauf) is a mutual assimilation, in principIe not unlike the Sanskrit case described earlier; but since both original segments are directly 'represented' in the output here, there has been no fusion: Sanskrit jauj-+ joj vs. English jauj-+ j:J:Jj. (ii) Stricture. The commonest type is opening ofstricture in response to surrounding opener stricture ('weakening', 8.3). So Spanish jb d gj-+ W O y] between vowels, Proto-Dravidian [c kj-+ [s xl in tervocalically in Tami!. Assimilation to closer stricture (,strengthening') is also attested, if rarely: in sorne southern U .S. English dialects, jzj-+ [d] before jnj ([bldnIs] 'business', [wxdnt] 'wasn't'), i.e. a fricative becomes a stop before a (nasal) stop. (iii) Lip attuude. The commonest type is rounding (usually anticipatory) of consonants in the vicinity of rounded vowels, e.g. [!,h] in English tore, to. Vowel rounding after rounded segments is also common; in Northumbrian Old English jej-+ [0] after jwj (Nhb woesa 'to be' vs. uiesan in other dialects); and a later revival shows up in those varieties of English with jwoj or jW:Jj for original jwaj, e.g. watch, wallet, swallow. (iv) Velic attuude. The commonest instance is anticipatory vowel nasalization before jN (C) [, as in most varieties of Englisb. This is the usual precursor to the development ofphonemic nasalized vowels: the deletion of a nasal afier it has nasalized a vowel lea ves [V]j[V] as a marker for what used to be [V]j[VN], thus phonemicizing vowel nasality. For example, French jftj fin vs. jfEj Jait, from Latin jfin-j, jfakt-j (plus other changes of course). Progressive nasalization occurs

8.2 Assimilation

and dissimilation too, ifless commonly: Sundanese appears to have a general rule nasalizing any sequence of vowels after a nasal, provided there are no intervening supraglottal articulations or boundaries: [maro] 'to halve', [jiin] 'to wet', [nj?s] 'ro take a holiday'. (v) Gloual state. Assimilatory voicing and devoicing are well attested, the former e.g. in Sanskrit voicing sandhi (4. 7), the latter in English external sandhi, e.g. [hzftu.] 'have to', [luesnu] 'has to' [ju:stu:] 'used to'. These are regressive; progressive voice assimilation occurs in the allomorphy of the Englisb plural, genitive, third person singular present and weak verbal past after obstruents: jsj in haioks, hawk's, tualks, jzj in bags, bag's, lags, jtj in uialked, jdj in lagged. Intervocalic voicing assimilation is also common, see the neutralization of the English jtj: jdj contrast (2 .8). Voice assimilation occurs in sonorants as well, with, in general, devoicing in response to voiceless obstruents. Thus Icelandic has progressive devoicing of [t l nj after jhj as in [hrirva] 'rake', [h10ypha] 'run ', [hl}i:vyr] 'knife', as well as regressive devoicing of these and jmj before (most) voiceless stops, as in [hsrp-pa] 'cassock', [VErpha] 'throw', [Eltha] 'pursue'; and progressive devoicing in final position after stops: [fYk'l] 'bird', [vohpn] 'weapon'. Glottal state dissimilation is rare, but there is an example in Maxakali nasal sandhi: [minni] 'black', [knnj] 'macaw', but [knniq rpinni] 'black macaw'. (vi) Complex assimilations. More than one parameter may be involved in assimilation; such processes characteristically result in the formation of geminates or identical c!usters from sequences of dissimilar consonants. Perhaps the most familiar examples are Latin preposition + verb compounds, like allero, auuli, alliitum 'carry to' (pres 1 sg, perf 1 sg, past part) , from ad-fer, ad-tuli, ad-ldtum, or colligii 'tie together' from con-ligo. A rather different case, involving nasals devoicing and becoming obstruents, occurred in early Scandinavian, where [-I)k-] -+ [-kk-], [-nt-] -+ [-tt-], [-mp-] -+ [-pp-]: Old Norse drekka 'drink', bau 'bound', kappi 'warrior' (original structures visible in O E drincan, hand, cempa). Acoustic assimilation Assimilation is normally thought of as articulatory adjustment (anticipation or persistence of vocal-tract configurations). But the concept can be usefully extended to cases where the trigger is an acoustic property, mediated via auditory perception. Since we have


ready alIowed for acoustic features atter to extend this to assimilation. anyhow (5.5), it is a simple

8-3 Phonological strengtlt

starts as a minimal coarticulation eflect is 01'ten sharpened or focussed ('phoneticized') into a perceptually more salient elTect, 01'ten leading to gross phonetic or phonemic change. Thus it looks as ir assimilation is bi-rnodal: there is the same 'choice' with regard to assimilation as there is with other phonological generalizations: we can respond either to articulatory or acoustic properties.

The folIowing sound change occurred erman:

9) i(:) ] [ e(:)

in some dialects 01' Austrian

[y(:) ]


!. unrounded front vowels round before velarized jlj. An articulatory Hure analysis would show no connection between rounding and any operty 01' [1); the change looks arbitrary. But acoustic analysis veals a different picture: velars (:md velarized segments) have neral 'Iow tonality' (specificalIy, low second and third formants); - rd the acoustic elTect 01' rounding is to lower these formants as well. 11'a segment with a low F2, say, folIows one with a high F2, the ansition between them wilI show a 'fiattening' or 'downshifting' 01' is formant. Imagine a speaker 'misreading' this downshift BETWEEN 'o segments, and locating the lower value in the first segment; and en altering the articulation 01' that segrnent to square with the .rceived elTect. SchematicalIy:

8.3 8.3.1



Lenition andjortition

Both synchronic and historical phonologies show frequent processes involving change both in stricture and glottal state. For various (not entirely obvious) reasons these are customarily grouped together, as follows: any movernent lO the right along the hierarchies in (8.11) below is lenition or weakening; any movernent to the left is fortition or strengthening:
(8. L 1) (a) Stop> Fricative > Approximant (b) Voiceless> Voiced > Zero

.10) [

High F2

11.0: F,J

Acoustic input

e' [ Low F 2

[LO: F,J

Speaker interpretation



Speaker output, as auernpt to conform to speaker interpretation

.etraction of[e:] to a central or back vowel would also be a reason,le response to the sarne acoustic input; on the indeterminacy 01' iimilarorv response see 8.6.) A similar instance occurs in many dialects 01' English, where nonvowels lower before a pharyngealized palato-velar jrj: here the 19ue position is in fact HIGH, but the transition is interpreted as a ver articulation 01' the preceding vowel, due to the high F value iociated with pre-pharyngeal transition. These transition effects, course, are maximized and extended in the course 01' time; what 5

The motivation is clear for (a): each step to the right increases the perrneability of the vocal tract to airflow. That for (b) is not imrnediately clear; but the frequency with which the change voiceless -+ voiced is a precursor to opening of stricture argues for an essential similarity; as do the coexistence 01' (a)-type and (b)-type changes as exponents of 'the same' process in languages. A case in point is the inidal rnutations. in Celtic languages, where in similar contexts (now morphosyntactic, once phonological), we have the following: (i) voiceless SlOpSbecorne voiced; (ii) voiced stops becorne fricatives; (iii) voiceless lateral ricatives (a) lose their friction, and (b) become voiced. Consider the isolative forms vs. rnasculine possessives of these nouns in North Welsh:
(8.12) Root N

(i) pen
(ii) braud (iii) <\01)

His N i bsn i vraud i 1::>1)

'head ' 'brother' 'ship'

Each initial goes one step down whatever hierarchy it's on; and jij goes from voiceless to voiced on (b) and fricative to approximant on (a). So 177

Phonological processes
phonologically the two are connecied; though the phonetic connection is obscure. (For a possible interpretation, see 11.4, 11.7.) But the traditional hierarchy (8.11) needs expansion: we ought to include affrication and aspiration as forms ofweakening as we/I, since they both involve opening 01' stricture in the release phase of a stop. Perhaps the best way to look at lenition/fortiuon overall is in terms 01' two strength scales, one ofopenness and one 01' sonority: rnovement down the first involves decreased resistance lo airflow, movement down the second an increase in the outpul of periodic acoustic energy. And a segment can, as we'lI see, move from one hierarchy to another. So:
Voicetess< 4a Aspirate> A ff' ricate .I Ora fricauve Gloual
fricative za ~

( r

8.] Phonological strength

then weaken (e.g. final [g] ~ [k] in German, and then in some dialects final rk] ~ [c,:]alter Iront vowels). The one place, however, from which fortition in the strict sense can't occur is zero: if a deleted segment is replaced by something, this is not a matter of strength any more. A word is in order on the sequence 3a ~ 2a in (8.13), which assigns a special status lO glottal Iricatives. This is in keeping with the c1aim in 6.5 that non-oral segments are 'defective', i.e. one submatrix short. There is strong evidence Ior this not only from synchronic processes like the Scots lel ~ [h] rule, but frorn the sources of Ihl in many languages. Perhaps the majority of Ihl in present-day languages can be traced back to the lenition of other obstruents; to take a few examples, all Germanic Ihl are from [x] frorn Indo-European Ik/; Armenian Ihl is frorn earlier Ip/; Latin Ihl is frorn earlier Ig/; Greek (Classical) Ihl frorn Is/; in Uralic, Ostyak, Hungarian, Yurak Ihl frorn Ik/; in Dravidian, Kannada Ihl from Ip/, Pengo, Kuvi Ihl from [e]; Manda, Kui Ihl from Ik/; and so on. In non e 01' these families can we reconstruct an original jhj. We further find that Ihl is particularly prone to loss; so much so that it can be regarded as a natural 'way-station' Irorn obstruent to zero. 011 this basis we can set up a two-phase progressive lenition schema for obstruents, divided into what we can call 'Ieature change' and 'rnatrix change' (alter Lass 1976a: 163):
(8.14) Progressive Lenition r: feature change [ - cont] ---> [ + cont] or [ --:: voice]
---> [

sonoriza tion


0 I

Voieed stop - ..5b

Af:rI icate b 4

l. --FricalIVt: 3 b

A1 iproximam zb




(8.13) defines a set 01' coordina les for strength-changes: down and/or right is lenition, up andjor left is fortition. Taking a voice/ess stop as the strongest segment type, we have two simple routes to zero: via opening alone, or via sonorization plus opening (or opening plus later sonorization). In general, movement to the right and down is commoner (more 'natural") than the opposite; and once (say over a language history) a segment has started on a path of lenition, it rarely appears to restrengthen. (This is not, as it may seem, a matter of lenition being 'cumulative' or 'directional': a segment does not know where it carne from, and if lenition is generally more probable than fortition, the odds are that any segment at any point on its hierarchy will weaken, if it does anything.) So input can be made at any point, and transfer can occur between sub-hierarchies, more or less at any point, e.g. a voiceless fricative can voice, and then continue down the opening scale at the (b) leve!. Converse/y, a segment may strengthen by moving from (b) to (a), and

Lenition [oral] [ [laryn]

11:matrix change

+ voice]

J [ J [0J

[laryn] (b)




Stage (c) of Lenition 11 may be called de-oralization, or dearticulation and produces a matrix type particular prone to loss. On the question 01' whether segments can 'skip' stages on the hierarchies, the evidence is ambiguous. Languages certainly exhibit ALTERNATIONS that look like progressions that skip steps (e.g. many Iorms 01' Northern English have [th] or [ts] ~ [J], as in [Jutsu?p] ~ [Jum?pl 'shut up' and the like); but it is unclear whether these substitution relations should be interpreted as processes in themselves, or as the relics of former historical processes, some of whose intermediate stages happen to be missing from the record. In the case

Phonological processes above, that would mea n that there ought to have been either a voiced stop or Iricative stage between the input and rJl. But keeping this issue on one side, we can illustrate various movements along the hierarchies, just to show the sorts 01" relations that come under the general heading 01" strcngth changes, and to illustrate the pervasiveness of'the pattems. As in previous sections of this chapter, 1 will use a rather unsorted mix of historical and synchronic examples. (In the material below, 'sa > 4a', etc. refer to the strength stages in (8.13).)
A Lenition


83 Phonological


(ii) Closing (1) eb > (3b?) > 4b: Latin f-j-f .... . talian fet/ (L maior, It maggiore); zb > 3b in French, majeur fma3cerf. (2) 3b > 5b: Southern U .S.fzf--' [d] ([bldl1ls] 'business', etc.). (3) zb > 5b (intermediatestages uncertain): lE *f-ww-, -jj-/--. f-gg-, -dd-f in some Germanic dialects: Old High German triuwa 'troth ', noeiio 'of two' vs. Old Norse lrygglla, tlleggja. (See any handbook of comparative Germanic under 'Holtzmann's Law'.) Preferential enoironments and '{notection' In a rather general way we can characterize particular environments as 'preferred' for certain strength changes. The notion of preference (just as in the choice of system elements, see ch. 7) is PROBABILISTIC: no environment is exclusively or predictably of one type or another, but certain processes occur so often in certain places that we can say 'X is a preferred weakening environment', etc. Judgements like this are basical!y inductive, i.e. they arise on the basis of extensive observation, though in some cases there is also a 'phonetic explanatiori' of sorts (see 8.6). For instance, V __ V is a prime weakening environment: al! things being equal, we expect lenition here. Thus the set of developments of Latin intervocalic lb d gl in the modero Romance dialects Iollows the expected picture o[ descent down the scales: (8.15) Latn -b-d-g1talian -v-d-


(i) Opening alone (1) 5a < 4a/4a': English /t/--. [th]/[ ts] (depending on dialect) initially; pre-Old High German /p t/--. [pfts] in certain environments (cf. Pfund, Her vs. pound, heart). (2) 5a> 3a > za: lE */k/--. Germanic /x/-+ [u]: lE */kerd-/ 'heart' --. Proto-Gmc */xert-/--.later /hert/ (E heart, G Hm:.); 2a> 1 pre-consonantally in English, */nokt-/ 'night', OE niht [nixt], 16th-century [neiht], Mod [nait]. (3) 3a> za > 1: Latin /f/--.Old Spanish /h/--. Mod /0/ ifilius- hijo /ixo/, inilial/h/ in 16th century). (ii) Sonorization alone (1) 5a > 5b: Proto-Dravidian */t k/--. Kannada /d g/ between vowels: /odagu/ 'help' from */lItakllj. (2) 3a > 3b: Old English /fe s/--. r v o z] between vowels; Welsh initial mutation ofvoiceless stops (see above). (iii) Sonorization + opening (1) 5a> 5b > 3b > zb: Latin /k/--.Old Spanish /g/-+ Mod Sp [y]--. Puerto Rican Spanish [w]: L aqua --. Sp agua --. PR /awa/. (2) 5a> (5b?) > 3b: Proto-Dravidian medial */t/-+ Tamil [o]: [uirui] 'drop off' rom */utir-j. B Fortition (i) Desonorization (1) 5b > 5a/3b > 3a: final devoicing of obstruents in German. The scale (8.13) does not have distinct places for voiceless nasals and approximants (indeed the status ofnasals is uncertain, since they do not typically weaken to anything except nasality on vowels, if thar's what that is): but with a suitable revision, the Icelandic devoicings in 8.2.2 (v) would come under this heading as well. (2) 3b> 3a: devoicing of [v z y/ in Amsterdam Dutch: /fri:s/ for Fries, Vries (fri:s/ vs./vri:s/ in the south), /sEin/ .jn 'his', /XOl/ goed 'good'.

French -v-




0 0

0 0

(L habre 'have', It aoete, Fr avoir, Sp haber; L crdere 'believe', credere, croire, creer; L legere 'read', legere, lire, leer.) Not al! segments have weakened, and the weakenings haven't gone the same distance in all the languages (see 8.3.3). But the pattern is clear. And on the basis of many such observations we would be entitled to assume ....even in the absence of any other evidence, such as the actually attested 'parent' language, Latin -- that Italian was the most 'primitive' o[ the three modero dialects, in the sense of being closest to an 'original' condition. The special status of intervocalic position ....and indeed something else about the nature o[ strength - is suggested by the development of SlOpSin different positions in U ralic. Consider the reftexes of (a) initial */p t k/, (b) intervocalic */p l k/, and (c) intervocalic */pp tt kk in four U ralic dialects:

Phonological processes
(8.16) Proto-Uralic *p*-p*-pp*t*-t*-tt*k*-k*-kk-

8-4 Whole-segment processes Finnish


Hungarian f~












More on strength hierarchies The U ralic data in (8.16) shows another interesting pattern. Ir you look at whieh segments weaken in particular languages, you note that there doesn't seem lO be - in any position - an across-theboard lenition: certain plaee categories are 'weak' or 'strong', and they vary rom language to language:







Hungarian Labial, velar Dental Labial, dental Velar


Ostyak Labial, dental, velar Labial, velar Dental

Velar Labial, dental Velar Labial, dental

Regardless of the failures of lenition, we notice the following: (a) if a stop weakens in initial position, it seerns to prefer opening to sonorization; (b) if a single SIOp weakens intervocalically, sonorization is usual, and opening frequeru; (c) the double SlOpSnever undergo either sonorization or opening, though they may shorten. Thus, if this material is representative, it suggests that both syllable-initial and intervoealic are weakening coruexts, but that the preferred modalities for lenition are opening initially and sonorization intervocalically. This in itself is not surprising, considering that if lenition is sornetimes al least an assimilatory response, V __ V should be more efTective as a trigger than __ V ('two vowels are more vocalic than one'). This is Iurther borne out by the resistance of Ippl etc. lO weakening: here one might say that two consonants are more consonantal than one, and hence double the resistance. Thus __ is a protected environment, i.e. consonants are not so prone lO lenition if protected by another consonant as they are standing alone. And this shows up in other, apparently unrelated cases, which can be made to fall together under a single generalization about protection. First, the failure of aspiration in most dialects 01' English for Ip t kl in clusters with Isl ([stek] vs. [thek]; in those dialects that do aspirate after Isl, the aspiration is usually shorter and less pronounced). Second, the failure of lE */p t kl lO become Germanic If x] if in c1usters with another obstruent: L spuii 'vomit', English speui, L stella, E star, L captus 'captured', OE 1/(4t 'captive' (here the protection works in one direction only - a preeeding obstruent proteets but is not itself proteeted, henee OE Iftl corresponding to Latin Ipt/). Third, in Greek Isl fails to become Ihl in clusters: hpta '7' (el'. L Jeptem) as expected, but esti 'he is' not *eht (el'. L est).



The strength c1asses also seem to be determined by difTerent features: Hungarian has [ + grave] as the weak initial cJass, [ + anterior] as the weak medial; while [ + grave] is the medial weak c1ass in Ostyak, and the strong initial in Vogul, ete. Data like this refutes the c1aim ofsome writers (e.g. Foley 1977) that eertain places are universally weaker than others (i.e. more prone to Ienition), and that there are implicational hierarehies such that if one c1ass lenites it will be velars, and lenition 01' dentals implies that 01' velars, ete. What is true, however, is that if at a given time a language has a weakening process, there is a strong tendency for certain place categories to be weak and others strong, in a given position. We ean add to the above the spirantization of intervoealic grave voieed stops in Old English (5.5), the U.S. English weakening of Itd/, the spirantization ol)t d k gl but not Ip bl in Liverpool English, and so on. At least for plaee, strength hierarehies are language-speeifie, not universal. We ean see the interaetion of universal and partieular strength relations iri a study by Zwieky (1972); he looked at the suseeptibility to deletion (as an index 01' weakness) of segments in one dialeet of English, and eame up with the following ranking:

Stops > Fricatives > 1) > m > n > 1 > r > h > w > j > Vowels are less

The types follow (8.13), to a large extent; but the positions independent. In the latter ease, 'strength' is a more arbitrary, phonetie coneept than it is for the larger cJasses. 8.4

Whole-segntent processes: insertion, deletion, reordering U p to now we have looked mostly at alteration of segmental


Phonological processes features; we have not heeu concerned in detail wirh lile crcation, deletion, or linear transposi tion of wholc segrncuts. Bu 1 proC<.:sscswith segment-size domains are cornmon and imporraut. Insertion New segmenls may appcar 'from zero' in ormerly unoccupied marginal posiiions in ihe word or morpheme, or betwcen two previously abutting segments. 'I'he general term or such insertion is epenthesis, Two types are worth defining separately, as the processes they denote often have special theoretical status. (i) Prothesis is the insertion oran initial segment, normally a vowel - usually with a phonotactic motivation. Thus in the transition between late Latin and Old French, initial Isp st skI clusters became illegal, and were destroyed by prothetic [e]: L spiritus 'spirit', stella 'star', scala 'Iadder', Fr esprit, toile, chelle (the two latter with later loss ol"/s/) . (ii) is the insertion ora vowel between IWOconsonants, most usually sonorants, or an obs + son or son + obs cluster. This is oten the sequel to syllabification of a sonorant: so in varieties of English with [Iilorn] 'film', [e9dli:t] 'athlete', the anaptyctic [d] may derive from earlier [film], etc. Anaptyctic vowels are also referred to as parasite vowels or by the Sanskrit terrn svarabhakti. There are also more 'general' epentheses that belong to neither of these types: vowel eperuhesis can occur in the forrn of diphthongization ofshort vowels, or (on a bi-segmental interpretation), lengthening of short vowels. Such epentheses may be assimilatory (as in [il-insertion after non-high vowels before palatals and velars in sorne U .S. dialects: S. Indiana [breig] 'bag', [boiJl 'bush', [bEiJ1tIJ 'bench'); or nonassimilatory, as in the shilt 01" earlier lel to ledl in certain Iorms 01" RP. Consonant epenthesis is equally common: one Irequent type sterns from a timing lag between a nasal consunant and a ollowing oral, where the velum closes before the 'target' post-nasal oral closure is formed. The result is an epenthetic oral stop, hornorganir to the nasal, but (usually) with the voicing value of the non-nasal. A familiar English example is the eperuhesis 01" [t] in I n,__ s] , [p 1 in Im_-fj/, and [k] in In__ 9/, as in [p.nnts] 'princejprints', [drsmpt] 'drearnr', [sAmp(11)] 'something', [IEl)k9j 'length'. Since all epentheses can be iruerpreted as 'replacement of zero' by sornething, the standard Iorrnalization is '0 ~ X'. Thus Frencli /e/1

8.4 Whole-segmet proeesses prothesis and English [dl-anaptyxis respectively


before sonorants

could be stated


0 ->e I





This 'works'; but stating the IwO in the same way rather misses the point, as does stating post-nasal stop epenthesis in the same format, e.g.
(8.20) +ObS - cont


[ [artic] P[ phon]

+ nas / [ a[artic] ]-[

+ obs PlPhonl]

(The conventions Ior agreernenl in place and phonation build 00 the discussion in 6.5; [a[ artic]] is a variable over al! possible values or place features, and [p[phon]] is equivalent to a variable over the laryngeal gesture: i.e. the epcnthetic stop agrees with the nasal in place and the following obstruent in phonation.) The point is that a process like (8. i qa) has a rather 'abstract' motivation, in terrns 01" a change in syl!able structure conditions; it is not phonetically motivated. Whereas in (8.lgb) and (8.20) there is a 'source' Ior the inserted segment. To clarify: anaptyxis typically occurs before or alter sonorants, which (vis-a-vis obstruents] are relatively more 'vocalic' or 'vowel-like'. From this we could argue for a feature [ + vocalic] shared by al! sonorants with vowels. Then we can interpret anaptyxis as seglDentalization or linearization: movement of a eature specification rom a position in a vertical column to a new place in a horizontal sequence. So for lefjli:tl ~ [efjdli:t]: ,
(8.21 )

- voc - voc ., . ., . ., . That is, rather than 'zero bccoming a vowel', a vocalic component is copicd out ami moved one place to the left, We can then assume that this moved componenl c1evelops a 'carrier', i.e. it is coordinated with sorne appropriate sct oplace Ieatures, derived either frorn the segment

+ obs
+ cons

- obs + cons + voc

+ obs ..... + cons - voc


l+ voc]


- obs

+ cons


Phonological processes
it's extracted Irorn (sce lxlow}, or lrom gClJeral vowl rondit ions in lhe stand

8.4 Whole-segment processes

language delining a 'mnima!' alone without other fcaturr-s This can be supported q ual i tativel y dependent derive

or 'neutra!' LO 'rmbody'

ti + VOl") can't

vowcls are

by languagl"s ofScots

in whirh

on Ica t LI res 01' consoua n ts thcy ea n be said lo

Irom. In sorne varicties

another sonoranr, rdl; alter

tlu-re is parasiting


1I1 ami
vowel is

(i) Aphaeresis is initial deletion: as in English 1 am -t l'm, 1 haoe I'VI', Gcrrnan une geh! es -t wie geht's 'how gocs it ', or the histor ical loss of init ial /k/ belorc /nl in English knife, kuight. (ii) Syncope (syncopation) is forrnative-internal delction: the 1 crm is most Ircq urn ti y uscd for vowel loss, bUI sorne wri ters ex rend it



as wcll. oAmerican 'secrcrarv',



see the


of syncopations words: Iseknteril




as injilll/,Ianl/,

but ih inscrted

cornparisons vs. /sckrltrl/

and British IdlkJdncri/

lorrns ofcertain process,

II( (whirh is velarized) it is rather back, and alter thc alveolar [v] it is rat hcr lronr: /lilAm], rfertm). Vocalicity and a (modified) place speciirauon llave becn ('xlractcd. We mighl forrnalize this using the formar t ypically used lor syntactic rransnot a 'neutral' forrnaiions (SO numbers identify (8.22) SD:

vs. IchkJI;1rI1 'dictionary',

etc, l t

may also show up as a systematic


as in a class of sonorant-



se = structure



final nouns in Swedish: tempel 'temple' vs. templet 'the temple', lager] lagre! 'camp', sdgenlsagnen 'legend'. (iii) Apocope (apocopation) is loss of a final elerneru. To take Swcdish again, in sandhi the final vowel of a nominal is also quite stem deletes


[-Ub' ]
+cons +voc IX back

beforc the plural

[- ohs ] + cons

sufTix: fiicka 'girl', for instance, consonant, are

flickor 'girls',




[ + voc IX back

man/rnen' 2 many beginning

(see 4. 7). Apocope 01" English, another deletions with

01' consonants


deletes before a word as in [last''atm] 'Iast lime'; low(as (8. [3) suggests) the last


stress words may also lose their Iinals, as in ami, o/ and velars
develops the sarne between in English could we can Historically, frequently a vocalic carrier, framework, and obstruents



be/ore And

palatals which using

be seen as copying here not responsive

out 01" [ lo place. moving

+ high]

siages of lcnitions; the weaker a segmeru is - in general - the more prone it is LO delerion. Thus consonants will increase in deletability as
rhey move generally able than right having stressed along lower the strength arnplitude

as a double (8.23)

the cornplex copying,







in two directions:

(8.20) is really:

+ nas

stressed ones, may be 'weak'

ones; and,

and/or perceptual salience than in a slightly dillerent sense) are more deleta slight extent, high vowels tend to be

["~:,7i:;J J
Thus we can distinguish

-conl] [ IX[artic

weaker than low, perhaps beca use 01' their inherenl relative shortness. But deletion is not always phouetically moiivated; many delctions are morphophonernic, seq uences ways while vowel ofa or across noun phonotactic, there a boundary, deleres beore or

[/J[~I~:~ J]t----Hf1I~~=~ J]
phonotacticallv or non-phonetically Irorn zero', rnot ifrom vated processes as genuine instances 01" 'segmcnl spurious eperuhesis or Ieature-segmerualization,



is clear

in The

insiance in Swedish
lwo-vowel dilerent siem-Iinal

vowel sandhi:

is a general

constraint involved. ending


bu t i t is instan tiated categories the plural


on the syntactic

(see 4.7,



thc initial vowel 01" the definite article deletes after a vowel-final noun:jlickalflickor,jfickaljlickan. So a negative condition * /V + V / operates as a kind 01" filter lO block ill-Iorrned strings that
would may otherwise have VI: 187 either may arise in the course apocope be said


] fsegmenls can emerge Irorn zero, they can also merge with it, i.e. delete. So the standard formal Ior deletion rules is the mirrorimage of that for epenthesis, i.e. 'X = a more specific traditional terrninologv: 186

01' derivation;

bUI this restriction The two to block

0'. As

or aphaeresis be involved

as its exponents, in a 'conspiracy'

with epenthesis,




IV +

Phonological processes

8-4 Whale-segment processes


Filter: */V + VI
Exponents: V


S 2


01 { __ -Jn

The formalism simply embodies the result, without


telling us how we


Reordenng Metathesis 01' transposiuon of segrnents is much less common than deletion 01' epenthesis, bUI occurs with some frequency as a historical change, and is occasionally found as an MP process. Most metatheses are sporadie (but see below or a systematie example). For instance, there have been a number 01' apparent metatheses in the history 01' English: thus in Old English we find interchanges 01' Ipl and Is/, as shown in spelling variants: Ips/--- spl in ioepse ~ wr.eJpe 'wasp', Isp/---/psl in epse r-c espe 'aspen', cosp r=cops 'copse', wlp~wlipj' 'Iisping'. (Note that the metathesized forms wasp, copse are now standard; though some dialects show wopse.) Another metathesis, 01' uneertain age, involves nasal sequences, particularly [tn] and [u]: emnity for enmity is quite frequent, and anenome for anemone seems to be developing near-standard status (judging Irom the frequency with which one hears even gardening experts using it). Note that these cases are all lexeme-specific, which is very common with metathesis: 1 have never heard *phomene, phenonemon, anemiiy, amaenia. A curious formal problem arises in the treatrnent 01' metathesis: should we interpret, say, Ips/--- Ispl as an 'interchange', or as a movement 01' one segment 'over the other'? And in the latter case, whieh one moves? Consider the possibilities:


got there. Now [or an example 01' systematic (morphophonemic) metathesis, where a segment interehange is part 01' a complex morphosyntactie process, interacting with other rules. Consider plural forrnation in Papago:

bana torna Ctho bahi

ba:bana ionona ichto ba:bhai

'coyote' 'knee' 'cave'


There are al leastthree

proeesses going on here: (a) reduplication 01' the initial syllable 01' the noun stem; (b) lengthening 01' the vowe\ in rhe reduplieated syllable if the original is short, and shortening 01' the stem vowel if the original is long; (e) metathesis 01' Ihl al'ter the reduplieated syllable. First, reduplieation: is this to be stated as 'syllable epenthesis'? Clearly not, sinee the motivation is morphological: the reduplieated sequenee should probably be taken as a realization 01' a plural morpheme, a phonological 'spelling' of an abstraet category. So:


PI +

V 2 + 3 + 3

01' the reduplicated


We can then adjust the quantity two rules:


syllable with
2 -+ V:

Original state (a)

(b) (e) p s s




V 5 V: 5




V 2






01' Ihl is then: SO: C

V 2 + 3

That is, Ipl can move to the right of /51 as in (b), Isl can move to the lefi 01' Ipl as in (c), or both as in (a). As far as 1 know there is no solu tion to this. The standard formulation for metathesis fudges the issue by not making a commitment. Thus the change above would be stated:


V 5

h 6



7 as a complement to the


a case of spurious


"(' Phonological processes spurious epenthesis discussed in 8+ l. The modern Ellglish forrns bright, fright are the result 01' an apparent metathesis 01' [v] and Ixl (OE h), as suggested by Iorrns like early OE be(u)rht,Iyrhto,Iorhtiga 'frighten', later breht,jrylzt,jrolztiga. Or they would be, if not for some interesting spellings aetually auesied. That is, we find variarus that can be arranged in an interesting sequence, like this:
(8.31) berht geberehtniga breht gebrehtniga fryht frohtiga 'bright' 'brighten' 'fiight' 'frighteri'

8.5 Complex processes and abbreoiatory notations expository devices, ways ofenabling us to get down on paper the kinds 01' siatements we want LO make, in the clearest way possible, with no particular theoretical status? 1f the former, then every notation is bound LO its theory, and we ha ve no right to use a notation outside its original framework, or with no reference to a particular theoretical context; if the lauer , we're entitled to use anything we need in order to express the descriptive generalizations we're after. 1 take the position that it is - at this stage anvhow rather siJly to claim that a notation is anything more than a visual metaphor, that it represents 'mental structure' or sorne aspect 01' an innate jacult de langage. Notational devices are distinct rom theories about what the notations may involve. The important point is that notations enable us to sharpen our perceptions of matters like rule-relatedness, and oficn to state (and even discover) generalizations that would be obscure or invisible without them (see lO.2). Phonological descriptions ideally aim for 'economy' - i.e. the avoidance of repetition, and the expression of generalizations in the sirnplest formo So for instance where rules involve partially similar structures, the notation should group the similarities together and throw the dissimilarities into relief. Current practice provides us with a host of notations that try to do just this. But it is important to note that however useful they are - their empirical foundation is obscure. Let us reiurn briefly to the question of'real' properties oflanguage being reflected in notations, One view is that particular abbreviatory devices that are applicable to linguistic data (obviously only a small subset 01' aJl conceivable ones) say something about the 'nature 01' language': a natural language is an object containing rules that can be abbreviated in certain ways, but not others. 11' we adhere to a strongly 'realist' /,mentalist' view, then it should be the case that the generalizations expressible in our notations are just those that speakers 'make use 01" in language-learning, processing, etc. (so Chomsky Ig65). This does not appear to be an



The 1-Vrx-I forms are historically older: ef. Gothie bairhts, farhts, farhtjan, which represcnt an older straturn 01' Germanie. What does (8.31) suggest? 1 think it suggests first of all that for the two missing forms, we ean fill in *berelzt, * geberlztniga. That is, the apparent metathesis, judging from the other forms, was preceded by anaptyxis ofa (more or less) matching vowel al'ter the [t], A reasonable scenario would be: (a) vowel-copy; (b) transler ofstress lo the copied vowel (no direct evidence, but the rest falls out neatly; we know tha t the vowel belore Ir 1 was origi nally stresscd ); (c) ddetion of the original, now unstressed vowel. That is:





CVrVxC 3


CrVxC 4

Given only and 4, we have metathesis; given representing either 2 or 3 (stress was not marked we have anaptyxis and syneope and (inferred) 'metathesis' as a sort 01' accidental side-efTect. 8.5

and a spelling in Old English), stress-shift, with

CODlplexprocesses and abbreviatory notations Any phonology involving processes will require notations that generalize over partly similar ones, and allow the formal unification 01' intuitively unified seis 01' sub-processes. The developrnent of phonological theory over the past two decades has been marked by, among other things, an increasing interest in such complex processes and their formalization, and the evolution 01' special formalisms for them. We have already introduced a number 01' these; in this section we will look at a few more relatively standard notations. The status 01' notations is rontroversial: are they basic elernents 01' the theory, standing for 'realities' 01' some sort:' Or are they simply

empirical issue. 1L is in fact not possible to produce a principled, non-intuitive justiflcation for abbreviatory devices like Greek-letter variables, braces, etc.; there is some marginal evidence from language history that rules MA y evolve so as Lo maximize abbreviability, but this is sparse (see Kiparsky Ig68b). For now, however, we can simply take it as a procedural imperative that within a process-phonology, abbrevi191

Phonological processes ation is desirable, and failure to use it where technically possible is a failure 01' analysis, 'rnissing a generalization'. The principie behind all abbreviation is primarily the avoidance 01' restatement; this leads to conHation, i.e. grouping 01' related rules into a scherna, 01' which they are subrules or expansions. Let us take an example in detail. 1n my speech, stressed vowels are nasalized before INC, and before a final nasal: [kh~:n?t) 'can't ', [kh~n) 'can'. So there are two partly identical rules:

8.5 Complex processes and abbreviatory notations From the LOpdown: a low non-back V lengthens before EITHERl (a) an obstruent which is EITHER2 [-cont, -t voice] ORz [+cont); OR1 before (b) a sonorant which is EITHER3 [ + nas, + ant] OR3 [ -Iat)' The unbracketed specifications in each larger matrix hold for the disjoined bracketed ones inside; each set 01" braces (as the subscripts above suggest) indicates another level of disjunction. Another form 01' disjunction allows insertion 01" subsidiary 'if-then' clauses inside larger ones (al! rules, obviously, are in one sense 'if-then' clauses). Consider the Old English sound change 'breaking ', which can somewhat oversimply be stated as:
(B.:)]) (a) Insert a [u] between a Iront vowel and a back continuant consonant. (b) But ir ihe back continuant is a sonorant, another must fol!ow.

+ stress





] (b) [V + stress

-> [

+ nas] 1__

[e ] e [e ] #
+ nas + nas


out similarities
[V ]

and enclosing the rest in braces:

-> [

+ stress

+ nasll

[e ] {e} #
+ nas

(For example l<ej---> [a-u] before [x] alone, but before alone: Ir 1/ were probably velar(izce!).) So the t\NOsubrules are:



Ir lI

(8.34) would informally be called a 'nasalization rulc' - but it is in Iact two rules conflated, i.e. a generalization over two processes that we want to claim are a 'unit'. What braces express, 01' course, is disjunction (see 5.5); but it is clear that we wouldn't really want LO speak ofitwo nasalization rules' here. Braces can also be used in more complex ways, 'nested' within each other, e.g. when a process aflects a scatter 01' only partly related items. They allow extraction 01' relevant contexts out 01' a larger whole. Thus the rule for lengthening 01' l<el in New York English runs (in pan) this way:
(835) l-.e/lengthens before: voiced stops, al! frica ti ves, al! sonorants except ID 1 and 1I1

0 ->

[~high ] + back


:'cont [ + back

] e


0 ->

[~high ] + back

e[~baCk ]

obs cont + back


A standard


would be:

[ ~ back ] -j Iow

-> [

+ long]


r{ [~:~~:, 1}l l J


We note one thing at the outset: the subrules must NOT apply in the order (a, b). Ifthis happened, then (a), which is more general (including al! back continuants) would allow [u ]-insertion before both obstruents 'and sonorants without a Iollowing C. We must have (b) apply first, since this is the more specific environment; then (a) can apply to what's left, which will be [ + obs]. We can handle this by abbreviating (8.38) in a particular way, with a l.ONGEST-FIRST convention on expansions. The subsidiary 'if-rhen' (8.38b) is expressed within angled brackets the first incorporating 'but if", the second 'theri'. Thus:

< ) .. < ),

I{ l [:::l}l [ J

(839) (b)


[~high ] + back



+ back +cont -obs)



iological processes longest-expansion-first on (8.38b-a).


8.6 Natural ensures the order of appli-


nother case where abbreviaiory notations appear to unify otherdisparate processes involves a spccial use of parentheses, along I an ordering condition as Ior angled brackets, and another, special Consider the basic rules or Latin accentuation: )
(a) Aceent the autepenulrirnate vowel ir the penultimate vowel is (i) shorr, (ii) in an open sylJable: hminis, denrius, regibus. (b) Otherwise aeeent the penultimare vowel, regardless oflength: ammuJ (penult long}, tmplum (pcnult closed). (e) Aeeent the single vowel in a rnunosylJable: Me, hs.

A schema like this requires a condition of disjunctive ordering: i.e. once a subrule has applied lO a given string, 110 other subrule may apply, even if its SD is met by the remaining string. 1'0 see why this is so, consider denrius: if subrules could apply wherever their SDs were met, we'd gel:
(8.44) (a) den~rius (b) den~rus (e) denhs (by (8.4Ia)) (by (8.4Ib)) (by (8.4Ie))

.e these three rules determine the local ion of all accents, there's a afacie case for generalizing: OUl how do we do it? First, to formalize subrules: 1)
(a) V->I+aee1/_eo[V (b) V -> r + aee (e) V->I+aCl.:l/ .se can be collapsed 2) -long V (:0 1/ J$eoVColI

One Iurther example of an abbreviatory uotation: the case of snirror-iInage envir onments. These are sequences where one SD is the reverse of anoiher, e.g. X --. Y I __ A, X --. Y I A __ . Thus some varieties of SCOlS have an allophonic rule where Itl (jish, hit) becomes [A] (identical to I Al in hut) ei ther before or after 11/, which is I!]' SO [hA!] = hulllhill, but the two vowels are distinct elsewhere. Collapsing with braces in the usual way, we would get: (Il.45)

1/ _


$ = sylJable boundary: see 10.3.5)

~ [

ront] - back 3 high


l+ back]

/ {[ + latl_} . __ [ + lat ]

with pareruheses

as ollows:
] $)

V -> [ + aee] / _

e, [V


e, V ) e, 11

We can gcneralize

this by suppressing

the environment-bar:

:ing the longest expansion Iirst (not excluding any material in nd brackets) we get (8'4Ia); iaking the second (excluding material he inner parentheses), we get (8'4Ib); and excluding all material .arentheses, (8.4Ic). Thus:
3) (a) d e n

(8.46) [


- baek 3 high

-> [ + baek] / [ + la t]

a + a

r e

u V



(Features based on Scots system in (7.38).) Not a very 'explanatory' rule as it stands: though ir we assume that [ + lat] --. [ + back] by redundancy rule, the FORM of (8.46) makes the point that it's the presence of [t] contiguous to a vowel - not its position - that counts. 8.6 Natural processes, evaluation sneasures, and explanation Consider these two rules:
(a) V ---> [ + nas] / _ (b) V


+ a

m e

u V



e e



[e ] + nas

[+ nas] / _


Phonological processes There is a clear dillerence: (SA7a) is 'reasonable' or expectable, (S.47b) 'arbitrary'. Or, pre-nasal nasaliz ation is phonetically lIlotivated, final nasalization unrnocivared. Phonetically motivated rules are generally - if loosely - referred lO as natural, unrnotivated ones as unnatural. In one strand of conlemporary thcory, the concept of naturalness (either as such, or in the form ofmarkedness) has been seen to interact with simplicity or economy, as follows: a noiation should be so designed that it SELECTS as simple, abbreviable, etc. just those aspects of phonology that are - in some sense- natural. Ami conversely, it should 'punish' (rather than 'reward') unnatural phenomena by making them harder to capture, more complex, cte. Now this can't be done by simple Ieature-counting (as had once been thought): in these terms, say, (8.47b) is 'simpler' than (S.47a). And we saw other instances of this in our discussion of natural classes (5.2, 5.5) - i.e. there are distinctly non-natural classes, like [ - cor], that are formally simple. But why should one want to quantify the notion ofnaturalness, and 'reward' and 'punish' analyses anyhow? 1lis not universally accepted that one ought to, but such atternpts have been made, and their failures are theoretically interesting. In the tradition 01' generative grammar, one strand, stemming frorn discussions in Chomsky ('g65) and elsewhere, has been the assumption that it is possible to construct, for a formalized theory, an evaluation lIleasure: a mechanical procedure that willjudge grammatical descriptions in such a way that a 'cost' is assigned to certain iterns, while others are 'free'. Thus any set of competing forrnulations can be judged as more or less 'costly', and the ideal is the 'cheapest' possible. In SPE an attempt was made LO bring together considerations 01' markedness and evaluation so that systems and rules could be assigned a cost on the basis of a set of lIlarking conventions. 1 will not go into detail on this matter, but just give an example of how the system is supposed LO work. Taking the general criteria or markedness set out in 7.4, we can say that the unmarked values lar roundness on vowels are: (a) [ - round] for front and low back vowels; (b) [+ round] for nonlow back vowels. Using u (unmarked) as a coefTicient, we can say: 8.6 Natural processes



1 ;-b~ck

l LI


I -+
[ - rounc\

L -Iow


11 [ + low J

(Where '__ ' inside a segment indicates a stmultaneous context, i.e. what is [u roundJ is a vowel that is la round, a back, -low] or round, + 10wJ.) We add the stipulation that in any rule or system display, an m (marked) value lar a eature has (say) a cost ofr and a u is cost-Iree. Thus given the systems li u 01 and li y u 01, the Iirst is free, and the second has a cost 01' r , (There are also costs for mid vowcls, non-back low vowels, etc.) But now observe this problem: it is apparently possible for a rule with a marked output LO be simpler than one wiihout:
(849) (a)


~ ~

- back v. +Illgh - round

-~ [ + back] (i





l '. J
-+ back


~ :'~:~~lc\J

+ rounc\

The SPE solution is to pro pose a concept oflinking, whereby when a Ieature changes in a rule, all other Ieatures whose markednessvalues depend on the changed feature take on their unmarked values with no cost. Thus since [+round] is linked to [+back, +high], (8.4gb) costs nothing - though forrnally it is more complex than (84ga). This is all very well, and elecrs the desired economy. The problem is that it appears, for all its ingenuity, to have no empirical consequences. That is: ifthere are cheap and expensive rules and systems, there ought to be some non-formal correlates to them, of the type suggested in 7-4- Take for instance the Germanic i-umlaut in (83): this anticipatory metaphony ought 10 be 'costly', since when lu 01 umlaut, they go to [y 0], thus incurring a cost 01' 2; linking would predicr the Iavoured output LO be [i e], which would have no cost. And


ological processes
.nly is the rule costly, but so is the resulting systern - at least afier narked vowels are phonologized (bccorne phonemic, no longer .hones 01' lu 0/: see 13.1). Assuming, that is, that markedness es are computed on a phonological, not a phonetic leve!. Here we a costly rule that overrides linking, and ultirnately a costly m: yet in the majority of the Germanic dialects, these rnarked -ls have remained stable Ior nearly a millennium and a half. .amples like this can be multiplied ad libitum. The problem is, Iy, that if an evaluation measure evaluates anything real (rather simply reftecting an irrational cross-linguistic distribution), there .t to be consequences assignable to marked states: instability, :ulty in learning, etc. And there is no real evidence that any such equences exist. fact, neither the idea of a formal evaluation measure in general, t quantifiable notion of naturalness/rnarkedness have made any y enlightening contributions - other than to the construction of ndex ofoddity' (see 7 .6.3). Nor do they tell us anything we don't idy know- though they do tell us things we don't want to know, use they're either untrue or unintelligible. the whole idea of naturalness then a dead end? Not necessarily, : put it into a reasonable perspective. Leaving asid e a purely .tical notion like rnarkedness, let's return to phonetic naturalness more transparent kind. lt's clear that phonetically natural esses are 'privileged' in a rather obvious way: (a) given a choice aatural vs. an unnatural process, the natural is much more Iikely; :iven the choice 01" no process vs. a natural one, there's a greater ihood ofthe natural process than no process (though (b) is a much cer predictor than (a) ). le question OfLlKELlHOODis the conceptual problem that bedevils .rrns oftheory which have a strong naturalness component. lf'both iatural and the unnatural occur, about the best you can say is that istance of a natural process is an occasion for lack of surprise, and rariwise for an unnatural one. But this doesn't rob the concept of ent: it makes it more suhtle and cornplex, and less straightarclly 'explanatory'. naturalness judgement is an answer to a 'why?'-question about ething. Why is (8.47a) natural? Because nasalization - all things g equal - is expectable be/ore a nasa!. Why? Because the velum ) fur an oral vowcl and down fur a nasal, and it's natural LO


8.6 Natural processes

anticpate an articulation. Why? Because speakers tcnd .. overall - to PREFERcertain articulatory configuraiions to others, Now ifwe examine these preferred types, it becomes apparent that many of them share a componenl which can (crudely but traclitionally) be callecl 'ease of articulation'. That is, perhaps the largest nurnber of natural processes are assimilatory: they tend in eflect to prolong particular gestures over larger stretches 01' phonic substance, or minirnize the nurnber of independent gestures in a sequence. This is wherc naturalness can becorne interesting. If a process is frequent, or a systern configuration 'unmarkecl', noting this is only the Iirst step in exploring its significance. The motivation (if any .- see below) will generally turn out to be either articulatory or perceptual; so markedness/naturalness are not explanations, but things to be explained. But -- ancl this is crucial+ it is a mistake to say that once you have discovered the motivation for a process, you have explained why the process occurs. You haven't: you've just said what it is about the process that makes it attractive to speakers who happen to 'want to achieve' the goal implied by the motivation. A simple exarnple will clarify: one of the cornrnonest 'natural processes' is nasal assimilation to following obstruents. The motivation is transparent: a cluster [nk], say, requires two gesture-shifts: one for the velurn and one for the tongue-body, while [I)k] requires only a shift 01" the velum. Therefore it's not surprising that if a speaker wants to minimize the 'effort' involvecl in NCj clusters, he will simply make the nasal homorganic with the stop. But, on the other hand, he coulcl do one of two other things: (a) make the stop homorganic to the nasal, i.e. [nk] ..... [nt]; (b) change the nasality and place values on the nasal, so that [nk] ..... [kk]; or (c) he could do a third thing, which is nothing al al!. Option (a) seerns virtually unattested; (b) ancl (c) are both common. So we now have sorne new questions: (a) why are both [nk] ..... [IJk] and [nk] ..... [kk 1 possible strategies [or 'resolving' heterorganic (nonhomorganic) INCj, but not [nk] ..... [nt]? Ancl why are there cases where [n k] is left alone? These are at present unanswerable, but some intcresting consiclerations arise frorn ihern. First, assimilation or any other natural process is subject to two problematical options: (i) 'Multiple strategy'. Given a potential assimilation, there is always more than one way to eflect a motivated change.

Phonological processes
Dor: hejdid hr .,il/.;:' Ile doesn'tididu' 'J'his rould In- treau-d as segmentalizing a fcature-complcx out of lhe auxiliary rompour-m , and attaching it 10 a n-l.uivcly 't:1l1pty' h-x ira l carrn-r (do lllight be callnl al! 'umpty veril' in much the sarnc way as [o] ea u he an 'cl11pty vowrl"). FuI' more on lineariz.u ion, Sigurd (1975). Ill:gatives without a lcxicu l On 'conspiracies', Kisscbcrth
(1 !:J7), Lass (1 <)74).

sillg vs. Can heit.ouh! he, l le can'tlcouldu't.

The limits of abstraction: generative phonology

B+2 B+3

Papago data alter Langacker (197\1). Fur discussion 01' the OE metatheses, Lass (1978). AIl intt'l't'stiug case 01' putative vowel rnerathesis is discussed in detail in SP1<-'(35~H1').

B.5 On rhe question 01' the 'realit y' 01' uouuions see SPH (ch. 1) vs. Robins (1957). The clairn that abbreviatory notations reflect real propert ies 01' language (and heuce, by a raiher devious argument, of'rnind') is made explicitly in Chomsky (1965: ch. 1); lor a sceptical discussion 01'this, and an interesting treauncnt of thc notion 'significant gencralization', H urlord (1t)77). On abbreviatory notations in general, **Harms (1968: ch. 7), **SPE (ch. Il). For disjuncrive ordering, SPE (34off). The discussion here has merely scrarchcd the surlacc 01'a very cornplex iopic. 8.6 For a detailed exposition ofthe issues raised herc, *Hymall (197.'): ch. 5). For the first modern markedness theory, **SPI!.' (ch. 9). For a derailed critique ofmarkedness, **Lass ('975); for a philosophical treatment of the issues, including the problem 01'naturalness aud explanation, * * Lass ('980: ch. 2). On naiuralness ser urther the * * Dressler/Lass exchange in Thrane el al. (1980: 75-' 0\1). On general properties of natural rules, Schane (1972). There is a school 01' thought ca\led 'natural phonology' (e.g. Starnpe 1969, '973, Donegan & Siampe 1977) which makes rule naturalness the prime consideration in theory, to the point 01' making a distinction bctween (innate) 'natural processes' and (Iearned) 'rules'. Their general view is that children are born with a ser 01' processes, and language learning consists at least partly 01' 'unlearning' ihese wired-in processes (like obstruent devoicing, vowel nasalization before nasals, etc.). For a critical view 01' this movernent, J. Ohala ('974), Householder (1977). This is not, by ihe way, to be coufused with 'Natural generarive phonology', which is a diflereru theoretical approach (9.6 below).

Tbe conceptual core: 'relation by mediation' We return in this chapier to more 'theoretical' eoneerns than those that occupied us in chs. 5-8: further away, anyhow, Irorn phcnornena that could (pretheoretieally) be said lo 'occur', and toward apparently deeper questions oftheoretical iruerpretation. We will be concerned with unobservables: items whose existence and naturc are detectablc only through the traces they leave, or inferrable by cornplex strategies of argument. Thc Iocus is a fundamental debate about the nature ofphonologieal represeruations and hence about the properties of the rule systems that relate thern lo phoneties. And this in turn brings us to the central issue: the accountability of phonology. Exactly what is it about, and where are ihe boundaries between it and other aspects of linguistic struciure? 'I'hese questions (already implicit in our discussion ofBloomfieldian process morphophonernics and the UUC, 4.3 6), can now be put more prccisely: how lar ought we lo lct (,deep') phonologieal represeruations diverge Irorn ('surl'ace') phonetic ones? And how Iar (if at all) should morphophonemics rather than allophonics determine the shape of a phonologieal description? Such questions are relevant particularly with rcspect lo the speeification of underlying forms, the segrnerus we allow to appear in thern, and the length, eomplexity, and ordering properties of the ehains of rules relating phonology to phonctics. What segrnerus, rnorpherne-shapes, etc. does a language 'really have' (sce 7'5,1, 7.7), and where do they appear? These issues, though irnplicit in all forms of 'process phonology', have been mOSI clearly Iocussed in the phonologieal theory evolved in the last 20 years or so in association with transformational grammar: so-called generative pbonology (GP). This is a developing set 01' 203