j anez Orešnik 
Ljubljana 
CDU 803.959 - 5-t 
FOUR MODERN ICELANDIC DEVOICING RULES 
Summary.1 Modern Icelandic generative phonology contains devoicing 
rules responsible for the partiaHy devoiced character of the final seg-
ments of words like dag, for the voicelessness of the stem final segments 
in words like dag-s, rusl(-s), and for the voicelessness of the segments 
followed by t in words like sval-t, skamm-t (the latter in southern pro-
nunciation only}. - One detail worth emphasizing: if the rules pre-
sented here are col'l'ectly formulated, the feature [voiced] is not distinc-
tive in the Modern Icelandic phonological segments /b d g/. 
1.1. The phonological component of Modern Icelandic grammar 
contains the following CONTINUANT DEVOICING rule: 
r- tense j (1) [ ] -+ [- voiced] / [ + syllabic] + continuant + [- voiced] 
l{- coronal} - anterior 
l.e. a:ny non-coronal or non-anterior lax continuant is devoiced if im-
mediately preceded by a syllabic segment and immediately followed 
by the morpheme boundary and a voiceless phonological segment. - At 
the point in the derivation when the Continuant Devoicing rule (1) 
applies, LAX (= non-TENSE) segments are simply short. The features 
VOICED, CONTINUANT, ANTERIOR, and CORONAL are to be un-
derstood in the sense of Chomsky and Halle 1968, see especially table (1) 
on pp. 176-77. For SYLLABIC, see ibidem, table (67) on p. 354; vowels 
are the only syllabic segments of Modern Icelandic . .....:... It follows from 
the list given here suh (2), and from the formulation of the Continuant 
1 My thanks are due to Miss Margaret G. Davis, who has improved the 
style of the paper. All errors are my own. The theoretical framework and 
the terminology of this paper are those of generative phonology as expounded 
by Chomsky and Halle 1968. Non-phonetic representations are bounded by 
the obliques, //, except in phonological derivations, where the obliques are 
omitted. 
137 
Linguistica XII 
(2) Lax voiced continuant phonological segments of Modern Icelandic: 2 
(a) coronal: /ct r 1/ (c) anterior: /v d 1/ 
(b) non-coronal: /v q/ (d) non-anterior: /q r/ 
Devoicing rule (1) that the rnle applies, in certain environments, to the 
Icelandic segments /v q r/, changing them to /f x r/, respectively.3 The 
reason for the very general formulation to the right of the plus will 
become evident below, in section 1.3. I am setting up rule (1) to account 
for the consonantal alternations in Modern Icelandic simplex words 
of the type exemplified in (3 a, b). That the application of the rule must 
bc limited to short /v r q/, can be seen from the examples given in (3 c-e). 
(3) 
(a) haf 'sea' 
vor 'spring' 
dagur 'day' 
(b) alcafur 'violent' 
akur 'acre' 
audugur 'wealthy' 
(c) bad 'bath' 
tal 'talk' 
heimur 'world' 
venja 'accustom' 
hringur 'ring' 
(d) hofud 'head' 
gamall 'old' 
atom 'atom' 
alinn 'fed' 
vikingur 'viking' 
(e) lcjarr 'thicket' 
ball 'dance' 
gramm 'gram' 
madur 'man' 
voiced stem final C 
gen. pl. haf-a 
gen. pl. vor-a 
gen. pl. dag-a 
n01n. sg. m. alcaf-ur 
dat. sg. akr-i 
n0111. sg. m. audug-ur 
gen. sg. bad-s 
gen. sg. tal-s 
gen. pl. tal-a 
gen. sg. heim-s 
gen. pl. heim-a 
voiceless stem final C 
gen. sg. haf-s 
gen. sg. vor-s 
gen. sg. dag-s 
gen. sg. m. akaf-s 
nom. sg. ntr. akaf-t 
gen. sg. alcur-s 
gen. sg. m. audug-s 
nom. sg. ntr. audug-t 
lp. sg. pres. ind. med. oen-st 
lp. sg. pres. subj. ven-j-i 
gen. sg. hring-s 
gen. pl. hring-a 
gen. sg. hofud-s 
gen. pl. hOfd-a 
gen. sg. m. gamal-s 
gen. sg. atom-s 
gen. sg. m. alin-s 
gen. sg. viking-s 
gen. sg. kjarr-s 
dat. sg. kjarr-i 
gen. sg. ball-s 
dat. sg. ball-i 
gen. sg. gramm-s 
dat. sg. gramm-i 
gen. sg. mann-s 
dat. sg. mann-i 
138 
j anez Orešnik 
The morpheme boundary posited in the structural description of the 
Continuant Devoicing rule (1) reflects the fact that the only segments 
which cause loss of voice in simplex words under rule (1) are /s/ and 
/t/, i.e. the only voiceless segments which can occur immediately to the 
right of the morpheme boundary in simplex words. (W ords such as 
sterkur 'strong', with voiceless l' in all their case forms, contain a phono-
logical voiceless /r/.) However, I am not aware of any compelling reason 
for the morpheme boundary to be obligatorily present in the structural 
clescription of the Continuant Devoicing rule (1). 
It is here stipulated that any segment to be devoiced by the Con-
tinuant Devoicing rule (1) be immecliately preceded by a vowel. This 
reflects the fact tliat rightmost voiced consonants in consonantal groups 
are not clevoiced by that rule. For examples of consonantal strings which 
do not unclergo rule (1), see (4 a), in which Einarsson's (1945) transcrip-
tions are presented. (On the basis of the situation in compound words, 
cf. section 1.3, Berkov-Bod:varsson 1962 can be assumed to concur.) A 
(4) Einarsson's transcriptions s.vv.: 
(a) horf 'direction' 
sfarf 'work' 
golf 'floor' 
kalfur 'calf' 
(b) torf 'sod' 
ulfur 'wolf' 
gen. sg. horf-s [-rvs] 
starf-s [-rvs] 
g6lf-s [-l(v)s] 
kalf-s [-l(v)s] 
torf-s [-rfsj 
ulf-s [-lfs] 
similar situation obtains in compound words, see section 1.3 below. On 
the other hand, Blondal 1920-24 devoices the rightrnost segments in 
all consonantal strings if they satisfy the structural clescription of the 
Continuant Devoicing rule (1), ignoring the [ + syllabic] segment of the 
structural description of the rule. Tlrns, the genitive singular forms 
analogous to those quotecl in (4 a) are all transcribed with [f] instead of 
[v]; see, for instance, Blondal's arf-s, s.v. arfur, ancl similarly in com-
pound words, cf. section 1.3 below. Einarsson has such transcriptions 
only seldom; for cxamples, see (4 b). I interpret this situation as inclicat-
ing that Icelandic has reformulated its Continuant Devoicing rule (1) 
since Blondal's tirne, limiting its domain to consonants immediately 
preceded by a vowel. (Einarsson's forms given suh (4 b) above are thus 
sporadic remnants of the olcler pronunciat.ion.) I assume that the change 
can be ascribed to the interaction of the Continuant Devoicing rule (1) 
and of the Cluster Devoicing rule (10), on which see section 2 below, 
2 Note that nasals are non-continuant in the distinctive feature system of 
Chomsky and Halle 1968. I-Iere and elsewhere in this paper the symbol q 
denotes a voiced velar continuant. 
3 Boldface indica tes voicelessness. 
139 
Linguistica XII 
but 1 am not able to describe the mechanism which has putatively 
caused the change. The matter deserves separate treatment. 
On the other hand, BlOndal 1920-24, Einarsson 1945, and Berkov-
Boctvarsson 1962 unanimously indicate, in their respective transcriptions, 
the devoicing of two consonants immediately preceding the morpheme 
boundary whenever the consonant immediately following the morpheme 
boundary is /t/ (/CC + t/, where both devoiced consonants pertain to 
the set {/v r q/}): e.g. nom./acc. sg. ntr. parf-t, with voiceless 1' and f, of 
parfur 'useful'. Moreover, the devoicing is indicated in the l of /lC + t/, 
where /C/ pertains to the set {/v r q/}, although so far no phonological 
rule is known which would devoice the l in such an environment; see 
the discussion of sjtilf-t in footnote 13. However, in no single case is the 
transcription with voiceless r/l and f the only one given; it is always 
accompanied by a transcription not containing the [f] intervening be-
tween the liquid and the t. Thus there is [part), which is a n9rmal form 
easily derived by aid of the Continuant Devoicing rule (1) after the 
deletion of /v/. 1 evaluate the situation just described as follows. The 
forms pronounced without [f] are normal; those pronounced with [f] 
and voiceless l are artificial. If this evaluation should prove wrong, my 
Continuant Devoicing rule (1) will have to undergo a major revision. 
1.2. In some cases the correct phonetic representations seem to depend 
on the assumption that the Continuant Devoicing rule (1) is preceded 
by some other phonological rule in the ordering. Two such cases will 
now be mentioned briefly suh (1-11). 
(1) The genitive singular of the noun bragd 'trick' is bragd-s, often 
pronounced [braxs]. Unless we are willing to see an exception in the 
latter form, its phonological representation must be /braqd + s/. (The 
velar cannot be voiceless, as it is on the phonetic level in the genitive 
singular, because it would then have to be voiceless in the phonological 
representation of the nom. sg. bragd as well, with the result that the 
Cluster Devoicing rule, discussed below, in section 2, would - wrongly 
- devoice the stem final d of this form.) To achieve the phonetic repre-
sentation [braxs], a rule deleting d between g and s has to be posited, 
and the derivation must be assumed to proceed as follows: 
(5) 
d-+0 / q-s 
Continuant Devoicing rule (1) 
braqct +s 
braq +s 
brax +s 
[braxs] 
Thus the ct-deletion rule must precede the Continuant Devoicing 
rule (1). . 
(II) The nom./acc. sg. ntr. of margur 'many' is marg-t [mart], with 
voiceless 1' and without g. Unless marg-t is allotted the status of an 
exception, for which there seems to be no need, its r cannot be under-
140 
] anez Orešnik 
lyingly voiceless in the framework of the present paper, for this would 
require the 1' to be voiceless in the phonological representation of, say, 
nom. sg. mase. margur; this, however, would lead to a wrong result on 
the phonetic level, for no rule has been posited which could voice the r 
in margur. Consequently the phonological representation of margt is 
/marg + t/, with voiced r. The derivation of margt proceeds as follows: 
(6) 
g-+0/r-t 
Continuant Devoicing rule (1) 
marg + t 
mar+ t 
mar+ t 
[mart] 
Since /g/ is not distinctively voiceless (cf. secti011 1.3), the Continuant 
Devoicing rule would not be able to devoice the 1' of margt if /g/ were 
still present in the representation when the Continuant Devoicing rule 
applied. Therefore it is necessary to postulate the ordering, the g-dele-
tion rule first, the Continuant Devoicing rule next. (The argument of this 
paragraph is valid only for the no;rthern form margt; in the homo-
phonous and synonymous southern margt, the T-Devoicing rule, dis-
cussed below, in section 4, may be responsible for the devoicing of the r.) 
1.J. The Continuant Devoicing rule (1) applies in compound words 
as well.4 See the illustrations in (7 a). That the rule applies only when 
the left constituent of the compound word ends in /v/, or /r/, or /q/, 
follows from the fact that other sounds than these are not devoiced in 
4 Within the framework described here, a voiced /v/ must be posited in, 
say, haf 'sea' on the phonological level. Not so in the comnound words like 
haf-sild '(kind of) herring'; here a phonological /f/ is not out of the question. 
True enough, a phonological boundary has to be posited between constituents 
of compound words in Modern lcelandic (Orešnik 1971 and footnote 15 below), 
and the only conceivable natura! motivation of this bonndary is achieved 
if at least (n-1) of the n constituents of any compound word are identified with 
some simplex in the lexicon. However, this identification need not be ex-
haustive. As is well known, simplexes can assume special forms when in-
corporated in to compound words as their constituents (Bloomfield 1933: 225, 
229). Thus it is conceivable that beside the stem /hav/ there is a compounding 
variant stem /haf/, and that the latter is used in haf-sild. Similarly, adfor 
'attack' when pronounced with [p] (as it sometimes is, see Bodvarsson 1963 
s.v. <l), and adferd 'method' when prnnounced with [p] (as it sometimes is, see 
Gudfinnsson 1946: 71), are not necessarily counterexamples to the Continuant 
Devoicing rule (1), because they can be assumed EITHER to contain a com-
pounding stem a]i- in the speech of those speakers who use the pronunciations 
just indicated, OR to have become simplex words through the loss of the 
boundary between the two constituents. (There may he even other pos-
sibilities.) Such examples could he easily multiplied. 
Stili, it is a fact that the compound-internal sandhi oheys the Continuant 
Devoicing rule, in that it does not allow segment clusters which are destroyed 
by the Continuant Devoicing rule in simplex words. 1 interpret this situation 
as an argument for the view that in the majority of cases the identity of the 
left constituents of compound words with some simplex words of the lexicon 
is exhaustive in the sense intended herc. 
141 
Linguistica XII 
the said position, see (7 b). It can be seen in (7 a) that the segments 
which cause devoicing are not limited to /s/ and /t/, as in simplex words. 
It must also be noted that the lax constituent initial b, d, g, although 
phonetically voiceless in the speech of most speakers, do not act as 
devoicing segments, see (7 c), and consequently cannot be voiceless 
phonologically. The implications of this situation for the phonological 
theory are as yet unclear to me. 
(7) (a) The final segment of the left constituent is phonetically voiceless: 
af-lcomandi 'descendant' 
f or-seti 'president' 
lOg-fnedingur 'lawyer', and many other examples 
(b) The final segment of the left constituent is phonetically voiced: 
sam-lcennd 'sympathy' 
gramm6f6n-plata 'record' 
bil-slys 'automobile accident' 
vid-Teoma 'touch', and many other examples 
(c) The final segment of the left constituent is phonetically voicecl 
although the constituent initial b, d, g are phonetically voiceless: 
haf-gola 'sea breeze' sam-band 'connection' 
ser-deilis 'especially' ein-gongu 'exclusively' 
dag-bole 'diary' til-boct 'off er' 
vid-b6t 'adclition' 
As far as the clevoicing of strings of voicecl consonantal segments is 
concerned, the situation in leH constituents of compound words is paral-
lel to that obtaining in simplex worcls, cf. section 1.1 above. Blondal's 
transcriptions are as predicted by the Continuant Devoicing rule (1) if 
the [ + syllabic] segrneut of the structural description of the rule is 
ignorecl; e.g. torf-palc 'sod roof' is transcribecl with [rf]. Einarsson's 
transcriptions only seldom follow Blondal's; torf-palc is transcribed with 
[rf] in Einarsson as well, whereas rnany other words, e.g. starf-semi 'ac-
tivity', contain [rv]. The transcriptions in Berkov-Bodvarsson do not 
indicate clevoicing in cornparable situatious at all; even torf-palc is 
transcribed wi th [ rv]. -
In one respect, however, the cornpouncl words do not follow the sim-
plexes: in the behaviour of constituent final strings such as /rv/ before 
constituent initial /tj. While the nom./acc. sg. ntr. djarf-t, of djarfur 
'daring', is transcribed with [r(f)t] in the three handbooks consulted, the 
compound djarf-tcelcur 'daring' is transcribed with voiced r in Blondal 
and Berkov-Boctvarsson. (There is no suitable example of the kind in 
Einarsson.) This supports my claim that thc transcriptions with [rft] of 
words such as djarft are spmious. 
142 
Janez Orešnik 
1.4. To some extent, the Continuant Devoicing rule (1) also applies 
to any word final segment not separated from the immediately following 
word by a phonetic pause. I base this statement on two relatively large 
publishecl samples of transcribecl connectcd specch, Malone 1923 ancl 
Bergsveinsson 1941, which I have investigated. The statistical results of 
this investigation are summarized in table (8).5 It can be seen there that 
/r/ obeys the Continuant Devoicing rale (1) in word final position when 
no phonetic pause immecliately follows it, and the next word begins 
with a distinctively voiceless sound (not b, d, or g!). The data on /v q/ 
are statistically insignificant, ancl caution is inclicated because of the 
fact that there is no instance of total clevoicing of /q/ in the two sources, 
whereas there are at least solitary examples of partially devoiced or 
even voiced /q/. On the other hancl, the sentence eg sagdi 'I saicl' is to be 
found transcribed in the bandbooks, always either with voiceless word 
final [x], or without any consonant at all in the left worcl (see, for 
instance, Einarsson 1945: 23 and Berkov-Boctvarsson 1962: 962). 
(8) The sounds mentioned in the present table occur in word final posi-
tion, are immediately followed by a distinctively voiceless word 
initial sound, and no phonetic pause intervenes between the two. 
(a) /r v q/: 
voiceless [r] 
partially devoicecl [r] 
voicecl [r] 
[f] 
partially devoiced [v] 
[v] 
[x] 
partially devoicecl [ q] 
[q] 
(b) other sounds: 
voiced sound 
partially devoicecl sonncl 
voiceless sound 
N umbe1· of examples in 
Malone 1923 Bergsveinsson 1941 
31 18 
l 1 
1 1 (long r) 
o 1 
o o 
o o 
o o 
2 o 
1 1 
18 13 
2 10 
o o 
5 Malone 1923 is a little difficult to interpret. Line d of his transcription is 
essential in the respect under consideration: if the column under the segment 
observed contains fig·ures 2 or 3 in line d, the segment is voiced; if the column 
oontains fignres 8 or 9 in the same place, the segment is voicelcss; a dot instead 
of a figure means that the seg·nient has the same specification of the fcaturc 
[voiccd] as the immediately preceding segment; a closing parenthesis between 
two symbols in line d mcans that thc seg·ment to the left ·of the parenthesis is 
partially assimilated in voice to the seg·ment immediately to the right of the 
parenthesis. 
The statistics on Bergsveinsson 1941 are based on his narrow transcrip-
tion, i.e. on line c of his texts I and II. 
143 
Linguistica XII 
In more than a few cases included in table (8) the critical segments 
are partially devoiced although we would expect them to be voiceless, 
or voiced. Since the acoustic diff erence between voiced/voiceless and 
partially (de)voiced sounds is difficult to hear, especially in quick 
speech, such examples are here not considered counterexamples to the 
Continuant Devoicing rule (1). 
An experimental field investigation of the devoicing in inter-word 
sandhi remains a desideratum. · 
2.1. The Continuant Devoicing rule (1) cannot account for the con-
sonantal alternations exemplified in (9), where # denotes a phonetic 
(9) 
vopn 'weapon' 
rusl 'rubbish' 
gutl 'dabbling' 
tagl 'tail' 
pukr 'secret dealing' 
f alm 'fumbling' 
uml 'mumbling' 
klifr 'climbing' 
voiced stem final C 
dat. sg. vopn-i 
dat: sg. rusl-i 
dat. sg. gutl-i 
dat. sg. tagl-i 
dat. sg. pukr-i 
gen. sg. f ftlm-s 
dat. sg. f alm-i 
gen. sg. uml-s 
dat. sg. uml-i 
nom. sg. klifr#1 
dat. sg. ldifr-i 
Dialect pronunciation with stop before d: 
sagdur 'said' nom. sg. m. sagd-ur 
haf dur 'had' nom. sg. m. haf d-ur 
voiceless stem final C 
nom. sg. vopn# 
gen. sg. vopn-s 
nom. sg. rusl# 
gen. sg. rusl-s 
nom. sg. gutl# 
gen. sg. gutl-s 
nom. sg. tagl# 
gen. sg. tagl-s 
n01n. sg. pukr#a 
nom. sg. f. sogd# 
nom. sg. f. hOf d# 
pause. Rule (1) cannot devoice the stem final segments in the forms of 
the rightmost column of (9), because the segments to be affected are 
either word-final or non-continuants and/or anterior AND coronal. To 
account for these voice alternations, I posit the following CLUSTER 
DEVOICING rule (10): , 
(10) [- syllabic]-+ [-voiced] / not [ + voiced] -{ ! [- voiced]} ~b~ 
6 The r of the nom. sg. pukr is totally voiceless, not only partially devoiced. 
My source of pukr, without epenthetic u, is Guc1finnsson 1946: 144. 
7 The r of the nom. sg. klifr is partially devoiced when followed by a 
phonetic pause. It is, however, not voiceless, as it should be if the Cluster 
Devoicing rule were not blocked by the presence of voiced [v] in the word. 
My sources for klifr, without epenthetic u, are Blondal 1920-24, Berkov-Boc1-
varsson 1962, Boc1varsson 1963, etc. 
144 
] anez Orešnik 
l. e. any non-syllabic segment is voiceless if immediately preceded by 
a segment that is not voiced, and immediate]y followed by the word 
boundary, or by the morpheme boundary AND a voiceless segment, in 
that order. - By "a segment that is not voiced" 1 mean any phonologi-
cally voiceless segment or any segment which is neither voiced nor 
voiceless at the tirne of the application of rule (10).8 /b d g/ are segments 
of the latter type. 
By way of illustration consider a few forms of rusl 'rubbish': the 
phonological representation of the stem is /rusl/, with voiced /1/. The 
voiced stem fina! segment is preserved before desinences beginning 
with a vowel, e.g. dat. sg. rusl-i. Since /1/ is preceded by a voiceless /s/, 
rule (10) applies in the nom. sg. rusl, where /1/ is word final, and in the 
gen. sg. rusl-s, where it is followed by the morpheme boundary and the 
voiceless desinence /s/, and changes II/ into the phonetically voiceless [l]. 
No such processes apply in, say, uml 'mumbling', for /1/ is here preceded 
by a voiced /m/. 
The following segments are devoiced by the Cluster Devoicing rule 
(10): /m n r 1/ and - in dialect - /ct/. For examples, see (9). 
In ali the examples known to me in which case (a) cif rule (10) ap-
plies, the devoiced segment is immediately followed, by the morpheme 
boundary. For this reason the morpheme boundary has been posited in 
(10 a). However, 1 am not aware of any compelJing reason for it to be 
there. 
Case (b) of the Cluster Devoicing rule (10) could be a part of the 
Word Final Devoicing rule (15), discussed below, in section 3. It would 
be natura! to say that voiced non-syllabic segments are totally devoiced 
if precedecl'by a segment that is not voiced, and only partially devoiced 
otherwise, in the enviromnent of the Word Final Devoicing rule (15). 
1 am not able to choose hetween these alternatives. Below 1 tacitly as-
sume, for purely practical reasons, that the Cluster Devoicing rule (10) 
is correctly formulated. 
. ' 
2.2. In some cases the correct phonetic representations seem to depend 
on the assumption that the Cluster Devoicing rule (10) is preceded by 
some other phonological rule in the ordering. Two such cases will now 
be briefly mentioned suh (1-11). 
(1) lf the phonological representation of the stem of f jall 'mountain' 
is /fjadl/, and if the alternation [dl,...., lj of the nom./acc. sg. fjall vs. gen. 
sg. f jall-s [fjals] is due to a d-deletion rule which deletes the /d/ in the 
segment group /dls/ whenever the three segme.nts involved pertain to 
8 No systematic status can he claimed for the ad hoc features NOT 
[ + VOICED] used in the formulation of the Cluster Devoicing rule (10), and 
[PARTIALLY VOICED] used in the formulation of the Word Final Devoicing 
rule (15), until more is known ahout the distinctive features of Modem lce-
landic consonantal segments. 
10 - Linguisticn 145 
Linguistica XII 
the same simplex word (contrast fjall-safn 'gathering of sheep in moun-
tains', with [dls]), then this rule must apply before the Cluster Devoicing 
rule (10), or else the phonetic result would be a wrong genitive singular 
form, (fjals], with voiceless l, for the derivation of the form would 
proceed as follows: 
(11) 
Cluster Devoicing rule (10) 
d-deletion rule 
fjadl +s 
fjadl +s 
fjal +s 
*(fjals] 
(II) The nouns fugl 'bird' and tagl 'tail' treat their consonant group 
gl alike in ali their respective case-forms except in the genitive singular, 
where fugl-s contains a spirantal g and a voiced l, whereas the g of 
tagl-s is a full stop, and its l is voiceless. One way to account for this 
difference in pronunciation is to posit different segment clusters in the 
phonological representations of the two nouns: /fuql/ vs. /tagl/. A phono-
logical rule (presumably the same rule which also changes /v q/ into 
stops before n, see Orešnik 1972) changes /q/ to /g/ before /1/ whenever 
the latter is not followed by a trne consonant. (The stipulation that the 
consonant be "true" is presumably necessary, for a following /r/ pro-
bably does not block the creation of the stop.) This leaves fugl-s intact, 
with (q], but creates /gl/ in the remaining forms of the word. lf this 
treatment of the difference in pronunciation between fugl-s and tagl-s 
is correct, the rule which changes /ql/ to /gl/ must precede the Cluster 
Devoicing rule (10), and the derivation of the case-form fugl proceed 
as follows: 
(12) 
ql-+ gl 
Cluster Devoicing rule (10) 
fuql 
fugi 
fugi 
(fygJj 
If the relative ordering of the two rules were the opposite, the Cluster 
Devoicing rule (10) would not have a chance to apply to /fuql/, seeing 
that the conditions for the application of the rule are not met by the 
latter representation: /q/ is distinctively voiced. This would leave us 
with no means of devoicing the /1/.9 
9 To account for the difference in pronunciation hetween fugi-s and tagl-s, 
two different lexical representations were posited in each case. The same 
treatment is necessary to account for the two pronunciations of bragd 'trick': 
the southern pronunciation is [braqd], with partially devoiced stem final sound 
before realized phonetic pause, the northern is [brakp] in the same environ-
ment; the case-forms in which the stem is followed by a vowel contain [qd] 
in the South, and [gd] in the North. The phonetically correct results are 
guaranteed if the lexical representation of the southern stem is /braqd/, 
and that of the northern stem fbragd/. The Cluster Devoicing rule (10) then 
applies in the northern nominative/accusative forms and devoices the word 
final /d/ into /p/. 
146 
Janez Orešnik 
2.J. As far as I can see, the situation in compound words is such as 
predicted by the Cluster Devoicing rule (10). I am basing this claim on 
the transcriptions in Einarsson 1945 and Berkov-Bi::id:varsson 1962. Sec 
(13). Bli::indal 1920-24, however, italicizes the final segment of the left 
constituent in the types of compound words exemplified by iungl-koma 
and jafn-gamall in (13); according to 6feigsson in Bli::indal 1920-24: XX, 
(13) environment of 
segment X voiceless 
final segment X of the left constituent 
voiced 
voiceless_voiceless vopn-fimi 'skill at arms' 
voiceless_ voiced 
voiced_ voiceless 
voiceless_/b d g/ 
fb d g/_ voiceless 
/b d g/_/b d g/ 
voiced_ voiced 
voiced_/b d g/ 
/b d g/_ voiced 
vopn-bitinn 'wounded w. arins' 
jafn-f adis 'on equal standing' 
jafn-gamall 'of the same age' 
vopn-laus 'unarmed' 
tungl-koma 'new Illl0on'10 
tungl-myrkvi 'lunar eclipse' 
tungl-braut 'lunar orbit' 
jafn-lyndi 'even temper' 
this means that the segments in question are sometimes pronounced 
voiced, and sometimes voiceless, but it is not clear whether they can 
vacillate in the speech of the same person, or of the same community, 
or anything else. The Cluster Devoicing rule (10) cannot account for 
this vacillation, and will have to be reformulated if the transcriptions 
of Bli::indal 1920-24 turn out to be nearer to reality than the more recent 
data on which this section is based. 
2.4. The handbooks offer hardly any information on the behaviour 
of the consonantal clusters enumerated in the left column of (13) in con-
nected speech. The Cluster Devoicing rule (10) predicts pronunciations 
such as those indicated in (14), where each pair of words is supposed 
to pertain to the same breath group. My own impressionistic observa-
tions of spoken Icelandic confirm the data presented in (14). However, 
a special investigation of this matter is a desideratum. 
A similar treatment helps to account for the two pronnnciations of blidka 
'soften': the southern with [pg], and the northern with [ctk]. /P/ and jd/, 
respectively, are posited in the lexical representation of the stem, and no 
devoicing rules apply in the derivations of the two pronunciation variants. 
Such simple treatment would not do in the case of the past participles hafdur 
pronounced with [b], of hafa 'have', and sagdur pronounced with non-continu-
ant g, of segja 'say'. Here underlying jhav + d/ and /saq + d/ must be posited, 
cf. the present-stem members of the respective verbal paradigms. A special 
northern rule changes /vd/ to /bet/, and /qct/ to /gd/, in the appropriate con-
texts. 
10 The pronunciation of tungl- which I have in mincl herc ancl in (14) cloes 
not contain [g]. 
147 
(14) environment of 
segment X 
voiceless-voiceless 
voiceless_voiced 
voiced_ voiceless 
voiceless-/b d g/ 
/b d g/ _ voiceless 
/b d g/_/b d g/ 
voiced_ voiced 
voiced-/b d g/ 
/b d g/ _ voiced · 
Linguistica XII 
final segment X of the left word is 
voiceless 
uopn fekk 
vopn beit 
jafn sfraumur 
jafn baggi 
voiced 
uopn let 
tungl kom 
tungl uex 
tungl gekk 
jafn lika 
J. In Modem Icelandic phonology there is also a WORD FINAL 
DEVOICING rule: . 
'[- syllabic l (15) + voiced ~ partially voiced / _ # 
- tense 
I.e. any lax voiced non-syJlabic segment is partially devoiced if im-
mediately followed by a word boundary. {See also footnote 8.) [ + voiced] 
must be mentioned in the structural description of the rule, otherwise 
distinctively voiceless segments, e.g. /s/, would be turned into partially 
voiced sounds by this rule. At the tirne that the Word Final Devoicing 
Tule (15) applies, lax non-syllabic segments are simply short. Examples 
(16) 
(a) bad 'bath' 
hef d 'title' 
dagur 'day' 
emj 'cry' 
ol 'ale, beer' 
horf 'direction' 
akur 'acre' 
gamall 'old' 
hofud 'head' 
gaman 'fun' 
talkum 'talcum' 
(b) skammur 'short' 
unna 'love' 
kjarr 'thicket' 
ball 'dance' 
stem final consonant 
voiced 
dat. sg. bad-i 
gen. sg. hef d-ar 
nom. sg. dag-ur 
dat. sg. emj-i 
gen. sg. ol-s 
dat. sg. horf-i 
dat. sg. akr-i 
dat. pl. goml-um 
dat. sg. hofd-i 
gen. sg. gaman-s 
gen. sg. talkum-s 
nom. sg. f. s ko mm# 
lp. sg. pres. ind. ann:jf 
nom. sg. kjarr# 
nom. sg. bali# 
partially devoiced 
nom. sg. bad# 
nom. sg. hefd# 11 
acc. sg. dag# 
nom. sg. em.i# 
nom. sg. ol# 
nom. sg. harf# 
nom. sg. akur# 
110111. sg. f. gOmul# 
nom. sg. hOfud# 
nom. sg. gaman# 
nom. sg. talkum# 
11 The pronunciation of hefd which I have in mind here contains [v] hefore 
ihe dental. 
148 
] anez Orešnik 
like hefd, emj and horf of (16) show that rule (15) must be limited to 
the word final position: voiced non-final non-syllabics in contact with 
word final segments are not partially devoiced by rule (15). 
The Word Final Devoicing rule (1.5) accounts for the consonantal 
alternations of the type exemplified in (16 a), where the simbol # de-
notes a phonetically realized pause. Examples showing that the rule 
must be restricte<l to short non-syllabic segments are given in (16 b). 
Einarsson 1945: 5, 25 asserts that the consonants in the endings -inn 
an<l -um retain some length and are voiced, at least in careful speech. 
These data are <lisputed. As regards unstressed -inn, they are stated 
similarly in Einarsson 1927: 38, 79, in Guctfinnsson 1946: 68 (where the 
author describes the length of [n) in the sai<l ending as vacillating 
between voiced long [n] and partially devoice<l short [n]), by implica-
tion also in Kress 1963, e.g. p. 57, as pointed out by Bencdiktsson 1965: 
112. On the other hand, Benediktsson l.c. believes that no such vacil-
lation exists, except perhaps in an affected lecture style of pronun-
ciation. As to the unstressed ending -um, the length of its m is assertecl 
by Einarsson 1927: 38, 79 and by Kress 1963: 33. Benecliktsson 1965: 112 
can be construed as denying the existence of any special length of m 
in -um. Whatever the truth about this quantity problem, the Word Final 
Devoicing rule (15) predicts that word final m ancl n will be voicecl when 
long, and partially clevoicecl when short. 
On the phonetic level, the eff ects of the Word Final Devoicing rule 
(15) can only be observecl at thc absolute end of breath groups, i.e. at 
phonetically realize<l pauses (Einarsson 1945: 24). Although the environ-
ment of rule (15) mentions the word boundary, ancl although the latter 
is present in the phonological representations of nouns with suffixed 
article (Orešnik 1972, app. B), and assumeclly present in the phonological 
representations of compound worcls (cf. footnote 15), the Worcl Final 
Devoicing rule (15) never seems to leave any traces behind in nouns 
with suffixed articles or in compound words. We return to this fact 
in section 4.3 below. 
4.1. The Continuant Devoicing rule (1), the Cluster Devoicing rule 
(10), ancl the Word Final Devoicing rule (15) apply with equal force 
in all Modern Icelandic dialects, as far as I know. This is not the case 
with the T-DEVOICING ruleto which we now pass. The T-Devoicing 
rule must be stated separately for southern and northern clialect areas:12 
(17) [-syllabic]-+ [-voicecl) / _t in southern pronunciation 
Le. in southern pronunciation any non-syllabic segment is devoicecl 
if immediately followecl by /t/. Rule (17) accounts for consonantal alter-
nations like those exemplified in (19). 
12 The terms SOUTHERN and NORTHERN are approximate labels. For 
a stricter geographical delimitation of the two pronunciations, see Gu(tfinns-
son 1964: 17-43. 
149 
Linguistica XII 
The northern counterpart of rule (1?) is as follows: 
(18) [ + lateral]-+ [- voiced] / _t in northern pronunciation 
Le. in northern pronunciation any .11/ is devoiced if immediately 
followed by /tj. Rule (18) accounts for the consonantal alternations 
exemplified suh (19 a). There are no comparable alternations in the 
words given suh (19 b), in northern pronunciation, hence the need for 
the more restricted northern T-Devoicing rule (18). The southern version 
(1?) is thus seen to be more general, and presmnably easier for language-
learning children to leam and remember, than the northern rule (18). 
This may be one of the factors contributing to the spread of (1?) at the 
expense of (18). 
(19) 
(a) svalur 'cool' 
(b) skemma 'damage' 
vanur 'used to' 
hringja 'ring' 
consonant before the hyphen 
voiced voiceless 
nom. sg. m. sval-ur 
inf. sl•emm-a 
nom. sg. m. van-ur 
inf. hringj-a 
nom. sg. ntr. sval-t 
supine sl•emm-t 
nom. sg. ntr. van-t 
supine hring-t 
Almost all the forms in which the T-Devoicing rule (1?/18) RAS to 
apply s·eem to involve /t/ preceded by a morpheme boundary at the 
point in the derivation when the T-Devoicing rule (1?/18) applies. (Words 
like vanta 'lack', with voiceless n in all the forms of the word in 
southern pronunciation, do not argue against the presence 0 f the mor-
pheme boundary in the structural description of the T-Devoicing rule 
(1?/18), for the voiceless n can be present in the phonological represen-
t~tion of the word, and is thus not necessarily dne to the T-Devoicing 
l'llle {1?/18).) I know only four exceptions: in the singular preterite in-
clicative of th~ strong verbs halda 'holcl', gjalda 'pay', svelta 'be hungry', 
and velta 'fall', voiced and voiceless l alternate with each other, as shown 
{20) (a) .voiceless l: lp. and 3p. sg. pret. ind., active helt, galt, svalt, valt 
(b) voiced l: 2p. sg. pret. ind. activ.e .and the whole sg. pret. ind. 
middle: helzt, galzt, svalzt, valzt 
in (20). No morpheme boundary can be posited before t in the forms suh 
(20 a). Nor can it be plausibly argued that the forms are suppletive 
forma:tions, with their voiceless l not due to the operation of the T-De-
voicing rule (1?/18), but 'present in the undedying representations of 
these forms, for one suppletive form would then be· necessary in the first 
and thircl persons singular preterjte indicative. active, and another in ali 
the remaining forms of the singular preterite indicative, including the 
middle voice. It seems to me much more plausible that the voicelessness 
150 
] anez Orešnik 
of the 1 in the forms of (20 a) is due to the T-Devoicing rule (17/18).13 To 
accommodate the forms sub (20), the rnorpheme boundary is not posited 
in the structural description of the T-Devoicing rule (17/18). 
4.2. Unlike the Continuant Devoicing rule (1) and the Cluster De-
voicing rule (10), the T-Devoicing rule (17/18) does not operate across 
the boundary between the constituents of compound words, or across 
the boundary between words.14 In fact, the morpherne boundary is the 
only boundary which does not block the T-Devoicing rule (17/18). For 
crucial examples in which the T-Devoicing rule (17/18) does not apply, 
see (21). 
4.'3. One reason why no attempt has been made here to collapse 
the T-Devoicing rule (17/18) with the Continuant Devoicing rule (1) 
and/or the Cluster Devoicing rule (10) is that the T-Devoicing rule 
(17/18) is, unlike the other rules just mentioned, a dialect dependent 
rule. Another argument against the collapsing is that the T-Devoicing 
rule precedes the Compound Boundary rule, whereas the Continuant 
Devoicing rule (1) and the Cluster Devoicing rule (10) follow the said 
boundary rule in the ordering. This matter will now be briefly discussed. 
13 The alternation between voiced and voiceless l is also observed in· the 
imperative singular of the verbs under discussion: the active form haltu con-
tains a voiceless Z, the middle form halztu a voiced Z. However, the T-Devoicing 
rule (1?/18) can account for this alternation even if its structural description 
contained an obligatory morpheme boundary, for the phonolog·ical represen-
tations of these forms are /hald + tu/ and /hald + st + tu/, where the mor-
pheme boundary before /tu/ may be a rewritten stronger boudary, in which 
case the rewriting rule operates before the T-Devoicing rule. 
The supines such as siglt of sigla 'sail' may also be relevant with respect 
tO the morpheme houndary in the structural description of the T-Devoicing 
rule (and, incidentally, even with respect to the structural descriptions of 
some other devoicing rules). However, these supines require separate treat-
ment hecause of the many prohlems associated with their derivation (meta-
thesis, etc.). 
Forms such as nom./acc. sg·. ntr. sjalf-t, of sjalfur 'self', are partly enigmatic. 
One of the p1:onu,nciations of sjalft is [-It], with voiceless l and without f; this 
pronunciation is easily accounted for by the T-Devoicing rule (1?/18) if the 
deletion of the phonological segment between l and t precedes the applica-
tion of the T-Devoicing rule; cf. footnote 18 ad finem. Another pronunciation 
of sjalft contains [-lft], with voiceless Z. This l cannot be devoiced by any of 
the rules posited here. I evaluate the [-lft]-forms as artificial. 
Guc1finnsson 1964: 1?-43, especially 30ff., reports some dialect pronuncia-
tions of sjalft. Normally his informants, school children, did not pronounce 
the f at all, and they either devoiced the lin accordance with the T-Devoicing 
rule in such forms (this was the normal situation), or did not devoice the Z; 
in the latter case · the T-Devoicing rule was only optional in their men tal 
grammars, or they (more seldom) lacked it altogether. In rare but interesting 
cases the f was pronounced as [v]: [-lvt], o.c. pp. 33, 35; this pronunciation is 
predicted hy ·my above devoicing rules for the dialect areas in which the 
northern variant of the T-Devoicing rule is in use. The Continuant Devoicing 
151 
Linguistica XII 
(21) (a) The final consonant of the left constituent is voiced: 
til-tala 'proportion' gamal-tungla 'barren old sheep' 
sam-tal 'conversation' at6m-takn 'symbol of chemical 
element' 
ein-tala 'singular' saman-tekning 'compilation' 
vid-tal 'talk' hofud-tilgangur 'chief goal' 
(b) The final consonant of the left word is voiced: 
bil tel bidil tel 
t6m tel at6m tel 
ein tel saman tel 
aud tel hOfud tel 
(Each pair of words suh (b) pertains to the same breath group.) 
On the phonological level words, simplex or compound, are bounded 
with word boundaries. There are also boundaries between the constit-
uents of compound words, although the precise nature of these bounda-
ries has not yet been determ.ined for Modem Icelandic; most likely they 
are word boundaries, and this assumption is accepted in the present 
paper.15 On the phonetic level, word boundaries are realized as phonetic 
rule (1) is blocked because the /v/ is not prececled by a vowel; the T-Devoic-
ing rule (18), which only affects /1/, cannot operate because of the intervening· 
/v/. Gudfinnsson reports cases of [-lvt] from the Northern districts only: from 
Eyjafjardarsysla (inclucling Akureyri) ancl Suctur-pingeyjarsysla. (The voiced 
pronunciation of the middle segment in [-lvt] cannot have been influencecl 
by the spellings in the text which the investigator asked the chilclren to 
read during the interviews; the text in question, as published by Gudfinnsson 
1946: 145-46, contains the forms ljuft and lift, which could have led to 
spelling pronunciations with [-vt] as well, but the author reports none.) 
To facilitate the exposition, I now state the pronunciations of sjalft 
predicted by my devoicing rules: 
with [f] or [v] 
without [f] or [v] 
South 
[-lft] 
[-lt] 
No'rth 
[-lvt] 
[-lt] 
The [-lft]-form is adduced in Kress 1963: 42; Kress has [telft] teflt, the 
supine of tefla 'play chess'. . 
Incidentally, the cases of the absent T-Devoicing rule, mentioned above, 
involve the northem variant (18). Diachronically, these are cases of rule loss, 
presumably due to the very narrow domain of the rule. It should be recalled 
that in contact with the southern variant (17) the northern rule (18) is giving· 
way; this must be another facet of the same phenomenon. 
14 One apparent exception to this claim is mal-tid 'meal', which is sometimes 
pronounced with a voiceless 1, see Blondal 1920-24 s. v„ Gudfinnsson 1946: 71, 
and Berkov-Boctvarsson 1962 s.v. This word can be accounted for in the same 
way as adfOr and adferd, see footnote 4 above. 
15 That there must be a boundary between constituents of compound 
words, is proved in Orešnik 1971, where, however, the nature of that boundary 
was not strictly determined, although is was shown that several phonological 
152 
Janez Orešnik 
pauses. (Phonetic pauses delimit breath groups.) Since the number of 
the word boundaries on the phonological level largely exceeds the num-
ber of phonetic pauses, the phonological component of Modem Icelandic 
(in fact, of every language) must contain a mechanism which cancels 
some of the phonological word boundaries during derivation. 1 imagine, 
maybe prematurely, that such a mechanism consists of rules interspersed 
among other phonological rules, deleting certain word boundaries or 
replacing them by weaker boundaries, most likely by morpheme bounda-
ries. It is assumed here that the Modem lceland1c phonological c01n-
ponent contains a mechanism whose duty is to replace word boundaries 
with morpheme boundaries, so that at the end of derivations only those 
word boundaries are preservecl which correspond to phonetic pauses. 
The rules of this mechanism - let us call them BOUNDARY RULES -
apply at different stages in derivations. One of the earliest boundary 
rules aff ects the word boundary between a noun and a suffixed clefinite 
article (Orešnik 1972).16 A later bounclary rule - let us call it the 
COMPOUND WORD BOUNDARY rule - replaces the word boundary 
between the immediate constituents of compound words with the mor-
pheme boundary. A still later bounclary rule - let us call it the INTER 
WORD BOUNDARY rule - replaces certain word boundaries between 
words with morpheme bounclaries, and thus creates what are to be 
realized phonetically as hreath groups. 
We already know that, while the T-Devoicing rule (17/18) can operate 
across a morpheme boundary, it cannot operate across any other houncl-
aries. On the other hand, case (a) of the Cluster Devoicing rule (10) 
operates freely across the worcl boundary hetween constituents of com-
pound words, and most prohahly also across the inter-word word 
phenomena of Modern Icelandic can be handled more satisfactorily if the said 
boundary is assumed to be a word boundary, than without that assumption. To 
the arguments of o.c. in favour of the worcl boundary between the constituents 
of compound words it can be adcled that one should think the boundary 
hetween a noun and a suffixecl article to be weaker than the boundary 
between constituents of compound words. As the boundary between a noun 
and a suffixed article has heen identifiecl with a word boundary (in Orešnik 
19?2), it is likely that the boundary between constituents of compound words 
is at least as strong as a word boundary; and since the phonological theory 
does not provide any stronger boundary than the word boundary, thc 
bounclary bctween constituents of compound words is likely to be identical 
with the word boundary. 
10 Through my negligence a minus sign has been omitted in the formaliz-
ed version of the Enclitic Boundary rule as printed in Orešnik 19?2: 29. 
The formulation is therefore repeated here: 
[ J-+ [a word boundary] / [_ segmentL [-a enclitic] 
Le. any number of subsequent word boundaries remains if immediately 
followed by a non-enclitic word, and are rewritten as a morpheme boundary 
if immediately followed by an enclitic word. 
153 
Linguistica XII 
boundary within breath groups. Case (b) of the Cluster Devoicing rule 
(10) and the Word Fina! Devoicing rule (15) operate only at the word 
boundary at the end of a breath group. The Continuant Devoicing rule 
(1) can operate across a morpheme boundary and across the word 
boundary between constituents of compound words, whereas it is unclear 
whether it can apply across the inter-word word boundary. 
A natura! wav to account for this situation is to assume that the 
devoicing rules (i), (10), (15), and (1?'/18) are interspersed among the 
boundary rules in such a way that the state of af fairs just described 
. follows as a consequence of the relative rule orderings posited. The 
orderings which accomplish just this are stated in (22), q.v. The generally 
(22) T-Devoicing rule (1?'/18) 
Compound Word ·Boundary rule 
Inter Word Boundary rule 
{
Cluster Devoicing rule (10) } 
Word Final Devoicing rule (15) 
Breath group boundary -+ phonetic pause11 
accepted conventions concerning rule orderings ensure the desired results. 
For instance, no word boundary is mentioned in the T-Devoicing rule 
(1?'/18), and the rule applies before the word boundaries between con-
stituents of compound words have been rewritten as morpheme bounda-
ries; this automatically ensures that the T-Devoicing rule (1?'/18) applies 
only in simplex words and in those constituents of compound words 
which are not themselves compound words. No word boundary is men-
tioned in the structural description of case (a) of the Cluster Devoicing 
rule (10). This means that the rule cannot operate across those word 
boundaries present in the representations at the time when rule (10) 
applies. Since, however, the word boundary between constituents of 
compound words and the inter-word word boundary had been rewritten 
as the morpheme boundary by the tirne rule (10) applies, the rule can 
operate freely in simplex as well as in compound words, and between 
words within breath groups. On the other hand, a word boundary is 
rnentioned in case (b) of the Cluster Devoicing rule (10) and in the Word 
Final Devoicing rule (15). As the Cornpound Word Boundary rule and 
the Inter Word Boundary rule have applied by the fone that rule (10) 
applies, the structural description of its case (b) is only met at the 
breath-group final word boundaries. Sirnilarly, at the tirne that the Word 
Final Devoicing rule (15) applies, all the word boundaries within breath 
groups; except those bounding the breath groups, have been rewritten 
17 The relative ordering; of the rules within the braces is at present unclear 
to me. The Continuant Devoicing· rule (1), not mentioned in (22), certainly 
follows the Compound Word Boundary rnle. Its ordering with respect to 
later boundary rnles is at present nnclear to me. 
154 
Janez Orešnik 
as morpheme boundaries; consequently no effects of rule (15) can be 
observed in compound words or in words not followed by a phonetic 
pause.18 
REFERENCES 
Benediktsson, Hreinn. Review of Kress 1963. islenzk tunga 6: 109-17. Reyk-
javik, 1965. 
Bergsveinss-0n, Sveinn. Grzmdfragen der isliindischen Satzphonetik. Phono-
metrische Forschungen, Reihe A, Band 2. Copenhagen and Berlin, 1941. 
Berkov, Valerij P„ and .Arni Bodvarsson. islenzk-russnesk ordab6k, Mos-
cow, 1962. 
BlOndal, Sigfus. islenzk-donsk ordab6k. Reykjavik, 1920-24. 
Bloomfield, Leonard. Language. New York, 1933. 
18 As the relation of the devoicing rules to the Enclitic Boundary rule 
mentioned above is not relevant to the problem under discussion, the matter 
has been relegated to this footnote. The T-Devoicing rule (17/18) FOLLOWS 
the Enclitic Boundary rule; hence, the Continuant Devoicing rule (1) and the 
Cluster Devoicing rule (10) follow it as well. For evidence that the T-Devoic-
ing rule follows the Enclitic Boundary rule in the ordering, consider the 
derivation of the supine rigni of rigna 'rain'. The derivation of the definite 
dat. sg. hring-num of hringur 'ring' is added for comparison; the word boundary 
and the morph /in/ are justified in Orešnik 1972, app. A, B: 
V owel Syncope rule 
metathesis 
nasal assimilation 
Enclitic Boundary rule 
b, d, g-+0/ _D, n 
T-Devoicing rule 
rign + D + t hriNg ##in+ um 
ring+ D + t 
riNg + D + t 
riN + t (2 applications) 
riN + t (southern only) 
luiNg ## n +um 
hriNg + n +um 
hriN + n +um 
(The /D/ in the phonological representation of rigni is phonologically a 
cover symbol for a dental consonant, and morphologically the dental suffix. 
Its presence in the phonological representation of the supine is not proved; 
however, the point made here remains valid even if the trne phonological 
representation of rigni is just /rign + tj. /N/ represents the velar nasal.) The 
last three Iines of the derivation are relevant to the point under considera-
tion. The relative ordering of the Enclitic Boundary rule and of b, d, 
g· -+ 0 j _D depends crucially on the question as to whether the deletion of g 
in rigni is performed by the same rule as the deletion of g in the definite dat. 
sg. hringnum. It is assumed here that this is the case. It is claimed in Orešnik 
1972 that the loss of g in hringnum follows the Enclitic Boundary rule. Hence, 
g·-+ 0 / _D follows the Enclitic Boundary rule as well. Since the T-Devoicing· 
rule (17/18} can only apply to the nasal of rignt after the deletion of g, the 
T-Devoicing rule must follow b, cl, g-+0 /_D. Hence, a fortiori, the T-Devoic-
ing rule must follow the Enclitic Boundary rule. Consequently the Continuant 
Devoicing rule (t) and the Cluster Devoicing rule (10) also follow the 
Enclitic Boundary rule. 
Incidentally, the derivation of rignt shows that the present formulation 
of the T-Devoicing rule presupposes that a number of other phonological 
rules precede it in the ordering and thus create the conditions for its opera-
tion: the metathesis rule; nasal assimilation; b, d, g-+0 /_D. Other rules 
which must precede the T-Devoicing rule include the loss of /v/ between /1/ 
and dental, cf. nom./acc. sg. ntr. hal(f)t of halfur 'half'. 
155 
Linguistica XII 
Bodvarsson, Arni, ed. islenzk ordab6k handa sk6lum og almenningi. Reyk-
javik, 1963. 
Chomsky, Noam, and Morris Halle. The Sound Pattem of English. New 
York, 1968. 
Einarsson, Stefan. Beitriige zur Phonetik der isliindischen Sprache. Oslo, 
1927. 
Einarsson, Stefan. lcelandic Grammar. Texts. Glossary. Baltimore, 1945. -
f have used the slightly revised fifth printing of 1967. 
Gudfinnsson, Bjorn, M ftllyzkur. Vol. I. Reykjavik, 1946. 
Gudfinnsson, Bjorn. Um islenzkan framburd. Mftllyzkur II. Edited and 
prepared for the press hy 6lafur M. 6lafsson and 6skar 6. Halld6rsson. 
Studia islandica vol. 23. Reykjavik, 1964. 
Kress, Bruno. Laut- und Formenlehre des Isliindischen. Halle/Saale, 1963. 
Malone, Kemp. The Phonology of Modem lcelandic. Menasha, Wisconsin, 
1923. 
Orešnik, Janez. "On the phonological boundary between constituents of 
Modem Icelandic compotmd words." Linguistica 11: 51-59. Ljubljana, 1971. 
Orešnik, Janez. "On the epenthesis rule in Modem Icelandic." Arkiv for 
nordisk filologi 87: 1-32. Lund, 1972. 
Povzetek 
šTIRI NOVOISLANDSKA RAZZVENITVENA PRAVILA 
Novoislandska generativna fonologija vsebuje štiri razzvenitvena pravila, 
ki povzročajo delno razzvenitev končnih glasov v besedah kot dag, popolno 
razzvenitev končnih glasov osnove v besedah kot dag-s, rusl(-s) in popolno 
razzvenitev glasov pred t v besedah kot sval-t, skamm-t (v zadnji samo v juž-
nem izgovoru). - Od nadrobnosti je vredno omeniti: če so tu predložena pra-
vila izrečena pravilno, fonološka oznaka [zveneč] ni razločevalna (distinktiv-
na) v islandskih fonoloških enotah /b d g/. 
156