Gujarati phonology

• Sanskrit's phonemic vowel length has been lost.^[1] Vowels are long when nasalized or in a final syllable.^[2][2]
• Gujarati contrasts oral and nasal, and murmured and non-murmured vowels,^[2] except for /e/ and /o/.^[3] See #Murmur section for more.
• In absolute word-final position, the higher and lower vowels of the /e ɛ/ and /o ɔ/ sets vary.^[3]
• /ɛ/ and /ɔ/ developed in the 15th century. when Gujarati developed into a distinct language from Old Gujarati (spoken by Gurjars in northern Gujarat and western Rajasthan).^[4]
• English loanwords are a source of /æ/.^[5]

• A fourth nasal phoneme is postulated for the phones [ɲ, ŋ] and the nasalization of a preceding vowel [Ṽ].^[7] Before velar and palatal stops, there is variation between these; e.g. [mɑ̃ɡʋũ]~[mɑŋɡʋũ] ('ask for'), [ɦĩtʃko]~[ɦĩɲtʃko] ('swing').^[8]
• Stops occurring at first members of clusters followed by consonants other than /ɾ, j, ʋ/ are unreleased; they are optionally unreleased in final position. The absence of release entails deaspiration of voiceless stops.^[8]
• Intervocalically and with murmuring of vowels, the voiced aspirated stops /ɡʱ, dʱ, bʱ/ have voiced spirant allophones [ɣ, ð, β]. Spirantization of non-palatal voiceless aspirates has been reported as well,^[8] including /pʰ/ being usually realized as [f] in the standard dialect.^[8]
• The two voiced retroflex plosives /ɖʱ, ɖ/ and the retroflex nasal /ɳ/ have flapped subphonemic allophones [ɽʱ, ɽ, ɽ̃]. The plosives /ɖʱ, ɖ/ are unflapped initially, geminated, and after nasal vowels; and flapped intervocalically, finally, and before or after other consonants.^[6] The nasal /ɳ/ is unflapped before retroflex plosives and intervocalically, and in final position varies freely between flapped and unflapped.^[7]
• /ʋ/ has [v] and [w] as allophones.^[9]
• The distribution of sibilants varies over dialects and registers.
  • Some dialects only have [s], others prefer [ʃ], while another system has them non-contrasting, with [ʃ] occurring contiguous to palatal segments. Retroflex [ʂ] still appears in clusters in which it precedes another retroflex: [spəʂʈ] ('clear').^[10]
  • Some speakers maintain [z] as well for Persian and English borrowings. Persian's /z/'s have by and large been transposed to /dʒ/ and /dʒʱ/: /dʒindɡi/ ('life') and /tʃidʒʱ/ ('thing'). The same cannot be so easily said for English: /tʃiz/ ('cheese'), which is rarely pronounced as /tʃidʒ/.
  • Lastly, a colloquial register has [s], or both [s] and [ʃ], debuccalized to voiceless [h]. For educated speakers speaking this register, this replacement does not extend to Sanskrit borrowings.^[8]

Phonotactical constraints include:

• /ɭ/ and /ɳ/ do not occur word-initially.^[2]
• Clusters occur initially, medially, and finally. Geminates occur only medially.^[2]
• Biconsonantal initial clusters beginning with stops have /ɾ/, /j/, /ʋ/, and /l/ as second members.^[11] In addition to these, in loans from Sanskrit the clusters /ɡn/ and /kʃ/ may occur.
  The occurrence of /ɾ/ as a second member in consonantal clusters is one of Gujarati's conservative features as a modern Indo-Aryan language. For example, languages used in Asokan inscriptions (3rd century BC) display contemporary regional variations, with words found in Gujarat's Girnar inscriptions containing clusters with /ɾ/ as the second member not having /ɾ/ in their occurrence in inscriptions elsewhere. This is maintained even to today, with Gujarati /tɾ/ corresponding to Hindi /t/ and /tt/.^[12]
• Initially, s clusters biconsonantally with /ɾ, j, ʋ, n, m/, and non-palatal voiceless stops.^[11]
• Triconsonantal initial clusters include /stɾ, spɾ, smɾ/ - most of which occur in borrowings.^[11]
• Geminates were previously treated as long consonants, but they are better analyzed as clusters of two identical segments. Two proofs for this:^[7]
  • The u in geminated uccār "pronunciation" sounds more like the one in clustered udgār ('utterance') than the one in shortened ucāṭ ('anxiety').
  • Geminates behave towards (that is, disallow) [ə]-deletion like clusters do.

Gemination can serve as intensification. In some adjectives and adverbs, a singular consonant before the agreement vowel can be doubled for intensification.^[13] #VCũ → #VCCũ.

The matter of stress is not quite clear:

• Stress is on the first syllable, except when it doesn't have /a/ and the second syllable does.^[14]
• Stress is barely perceptible.^[15]
• Stress typically falls on the penultimate syllable of a word; however, if the penultimate vowel in a word with more than two syllables is a schwa, stress falls on the preceding syllable.^[16]

Schwa-deletion, along with a-reduction and [ʋ]-insertion, is a phonological process at work in the combination of morphemes. It is a common feature among Indo-Aryan languages, referring to the deletion of a stem's final syllable's /ə/ before a suffix starting with a vowel.^[14]

This does not apply for monosyllabic stems and consonant clusters. So, better put, #VCəC + V# → #VCCV#. It also doesn't apply when the addition is an o plural marker (see Gujarati grammar#Nouns) or e as an ergative case marker (see Gujarati grammar#Postpositions).^[17] It sometimes doesn't apply for e as a locative marker.

Between a stem ending in a vowel and its suffix starting with a vowel, a [ʋ] is inserted.^[19] #V + V# → #VʋV#. This can be seen in the formation of passive and causative forms of verb stems.

The second example shows an ɑ-reduction as well.

ə finds itself inserted between the emphatic particle /dʒ/ and consonant-terminating words it postpositions.^[20]