Icelandic has an aspiration contrast between plosives, rather than a voicing contrast, similar to Faroese, Danish and Standard Mandarin. Preaspirated voiceless stops are also common. However, fricative and sonorant consonant phonemes exhibit regular contrasts in voice, including in nasals (rare in the world's languages). Additionally, length is contrastive for consonants, but not vowels. In Icelandic, the main stress is always on the first syllable.
Even the number of major allophones is subject to some dispute, although less than for phonemes. The following is a chart of potentially contrastive phones (important phonetic distinctions which minimally contrast in some positions with known phonemes; not a chart of actual phonemes), according to one analysis (Thráinsson 1994):
A large number of competing analyses have been proposed for Icelandic phonemes. The problems stem from complex but regular alternations and mergers among the above phones in various positions.
Examples of alternations across different positions:
Voiced consonants are devoiced word-finally before a pause, so that dag ('day (acc.)') is pronounced ['ta:x], bauð ('bid (1/3 pers. sg. past)') is pronounced ['poe:i?], and gaf ('gave (1/3 pers. sg.)') is pronounced ['ka:f]. Even sonorants can be affected: dagur ['ta:r?] ('day (nom.sg.)'), ketil ['c:t?l?] ('kettle (acc.)') 
The "glottal fricative" (actually a placeless approximant) only occurs initially before a vowel, and following a vowel in the sequences [hp ht hk hc]. These latter sequences are sometimes said to be unitary "pre-aspirated" stops; see below.
The voiceless velar fricative occurs only between a vowel and or , and initially as a variant of [k?] before [v]. Because it does not contrast with [k?] in either position, it can be seen as an allophone of /k?/. However, it also alternates with [?], occurring before a pause where [?] would be pronounced otherwise.
There are two sets of palatal sounds. "Alternating palatals" [c c? j] alternate with the velars [k k? x ?], while "non-alternating palatals" [ç j] do not. Note that [j] appears twice here; these two [j]'s behave differently, occur in different distributions, and are denoted by different letters (g and j). This suggests that they may belong to different phonemes, and that is indeed a common analysis.
In general, the alternating palatals [c c? j] are restricted to appearing before vowels. Velars [k k? x ?] are restricted to appearing everywhere except before front vowels. In other words: Before back vowels and front rounded vowels, both palatals and velars can appear; before front unrounded vowels only palatals can appear; before consonants only velars can appear.
For the non-alternating palatals [ç j]: Both can appear at the beginning of a word, followed by a vowel. Elsewhere, only one can occur, which must occur after a non-velar, non-palatal consonant. [j] occurs before a vowel, and [ç] occurs in a few words at the end of a word following [p t k s].
The velars and alternating palatals are distributed as follows:
Although the facts are complex, it can be noticed that only ever contrasts with one of the two velar stops, never with both, and hence can be taken as an allophone of whichever one doesn't appear in a given context. Alternatively, following the orthography, [?] can be taken as an allophone of , where is taken as an allophone of either or depending on context, following the orthography.
In native vocabulary, the fricatives and are allophones of a single phoneme . [?] is used morpheme-initially, as in þak ['?a:k] ('roof'), and before a voiceless consonant, as in maðkur ['ma?k?r?] ('worm'). [ð] is used intervocalically, as in iða ['?:ða] ('vortex') and word-finally, as in bað ['pa:ð] ('bath'), although it is devoiced to [?] before pause. Some loanwords (mostly from Classical Greek) have introduced the phone [?] in intervocalic environments, as in Aþena ['ana] ('Athens').
Of the voiceless sonorants [l? r? n? m? ], only [l? r? n?] occur in word-initial position, for example in hné ['n?j?:] ('knee'). Only in initial position do the voiceless sonorants contrast with the corresponding voiced sonorants. Finally, before aspirated consonants and after voiceless consonants only the voiceless sonorants appear; elsewhere, only the voiced sonorants appear. This makes it clear that [m? ] are non-phonemic. Recently, there has been an increasing tendency, especially among children, to pronounce initial hn as voiced, e.g. hnífur ['niv?r?] ('knife') rather than standard ['n?iv?r?].
The palatal nasals [ ?] appear before palatal stops and the velar nasals [ ?] before velar stops; in these positions, the alveolar nasals [n? n] do not occur. [?] appears also before [l], [t] and [s] through the deletion of [k] in the consonant clusters [?kl] [?kt] [?ks], and through the coalescence of the consonants [k] and [n] in the consonant clusters [knl] [knt] [kns]. The palatal nasals are clearly non-phonemic, although there is some debate about due to the common deletion and [k] coalescence of [kn].
Modern Icelandic is often said to have a rare kind of stops, the so-called pre-aspirated stops [?p ?t ?c ?k] (e.g. löpp ['loe?p] 'foot'), which occur only after a vowel and do not contrast with sequences [hp ht hc hk] (which do not occur in Icelandic). Ladefoged & Maddieson (1996) note that phonetically, in Icelandic pre-aspirated stops the aspiration is longer than in normal post-aspirated stops, and is indistinguishable from sequences [hp ht hc hk] (or with replacing ) occurring in other languages; hence, they prefer to analyze the pre-aspirated stops as sequences. For example, Icelandic nótt, dóttir correspond to German Nacht, Tochter.
Following vowels there is a complex alternation among consonant length, vowel length and aspiration. The following table shows the alternations in medial and final position (Ladefoged & Maddieson 1996):
'small pot' (nom pl)
'doorway' (nom pl)
'young seal' (nom pl)
'endow' (2nd p. sg past)
'dampness' (obl sg)
'can' (2nd p. sg past subj)
'careful' (masc nom pl)
'wide' (neut sg)
'dark' (fem nom sg)
'dew' (nom sg)
'bite' (1st p. sg pres)
'grasps' (nom pl)
In most analyses, consonant length is seen as phonemic while vowel length is seen as determined entirely by environment, with long vowels occurring in stressed syllables before single consonants and before certain sequences formed of a consonant plus [v r j], and short vowels occurring elsewhere. Note that diphthongs also occur long and short.
As discussed above, the phones [m? ? x ð], probably , and debatably are non-phonemic. Beyond this, there is a great deal of debate both about the number and identity of the phonemes in Icelandic and the mapping between phonemes and allophones.
There are a number of different approaches:
Phonetic vs. orthographic:
Maximalist vs. minimalist:
The main advantage of the phonetic approach is its simplicity compared with the orthographic approach. A major disadvantage, however, is that it results in a large number of unexplained lexical and grammatical alternations. Under the orthographic approach, for example (especially if a minimalist approach is also adopted), all words with the root sag-/seg- ('say') have a phonemic /?/, despite the varying phones [k], [x], [?], [j] occurring in different lexical and inflectional forms, and similarly all words with the root sak- ('blame') have a phonemic /k/, despite the varying phones [k], [k?], [hk]. Under the phonetic approach, however, the phonemes would vary depending on the context in complicated and seemingly arbitrary ways. Similarly, an orthographic analysis of three words for "white", hvítur hvít hvítt ['k?vi:t?r] ['k?vi:t?] ['k?viht] (masc sg, fem sg, neut sg) as /kvit?r/ /kvit/ /kvitt/ allows for a simple analysis of the forms as a root /kvit-/ plus endings /-?r/, /-/, /-t/ and successfully explains the surface alternation [i:t] [i:t?] [iht], which would not be possible in a strictly phonetic approach.
Assuming a basically orthographic approach, the set of phonemes in Icelandic is as follows:
The parentheses indicate phonemes present in a maximalist analysis but not a minimalist analysis.
There is a particular amount of debate over the status of and [c?]. A maximalist analysis sees them as separate phonemes (e.g. and , respectively), while in a minimalist analysis they are allophones of and before front unrounded vowels, and of the sequences /kj/ and /?j/ before rounded vowels, in accordance with the orthography. The maximalist approach accords with the presence of minimal pairs like gjóla ['co:ula] ('light wind') vs. góla ['ko:ula] ('howl') and kjóla ['c?o:ula] ('dresses') vs. kóla ['k?o:ula] ('cola'), along with general speakers' intuitions. However, the minimalist approach (e.g. Rögnvaldsson 1993 harvnb error: no target: CITEREFRögnvaldsson1993 (help)) accounts for some otherwise unexplained gaps in the system (e.g. the absence of palatal/velar contrasts except before rounded vowels, and the absence of phonetic [j] after velars and palatals), as well as otherwise unexplained alternations between palatals and velars in e.g. segi ['s?i:j?] ('[I] say') vs. sagði ['sa?ð?] ('[I] said'; assuming that [j] and [?] are taken as allophones of palatal and velar stops, respectively). On the other hand, the number of such alternations is not as great as for stop vs. fricative alternations; most lexical items consistently have either velars or palatals.
The voiceless sonorants are straightforwardly taken as allophones of voiced sonorants in most positions, because of lack of any contrast; similarly for vs. . On the other hand, [l? r? n? ç] do contrast with [l r n j] in initial position, suggesting that they may be phonemes in this position, consistent with a maximalist analysis. A minimalist analysis, however, would note the restricted distribution of these phonemes, the lack of contrast in this position with sequences [hl hr hn hj] and the fact that similar sequences [kl kr kn] do occur, and analyze [l? r? n? ç] as /hl hr hn hj/, in accordance with the orthography.
The velar nasal is clearly an allophone of before a velar stop. When it occurs before or as a result of deletion of an intervening , however, some scholars analyze it as a phoneme /?/, while others analyze it as a sequence, e.g. /n?/.
There is less disagreement over the vowel phonemes in Icelandic than the consonant phonemes. The Old Icelandic vowel system involving phonemic length was transformed to the modern system where phonetic length is automatically determined by the syllable structure. In the process of eliminating vowel length, however, relatively few vowel distinctions have been lost, as the loss of phonemic length has been offset by an increase in the number of quality distinctions and diphthongs.
|Mid to close||ei o oei||ou|
|Open to close||ai||au|
/ai/ has a front onset, , while /au/ has a back onset, .
Before other consonant clusters (including the preaspirated stops [hp ht hk] and geminate consonants), stressed vowels are short. Unstressed vowels are always short.
An exception occurs, if there is a t before the infix k. Examples are e. g. notkun and litka. There are also additional exceptions like um and fram where the vowel is short in spite of rules and en, where the vowel length depends on the context.