Summary

As was mentioned above, the dialects of Portuguese can be divided into two groups, according to whether syllable-final sibilants are pronounced as postalveolar consonants /ʃ/, /ʒ/ or as alveolar /s/, /z/. At the end of words, the default pronunciation for a sibilant is voiceless, /ʃ, s/, but in connected speech the sibilant is treated as though it were within a word (assimilation):

• If the next word begins with a voiceless consonant, the final sibilant remains voiceless /s, ʃ/; bons tempos [bõʃ ˈtẽpuʃ] or [bõs ˈtẽpus] ('good times').
• If the next word begins with a voiced consonant, the final sibilant becomes voiced as well /z, ʒ/; bons dias [bõʒ ˈdiɐʃ] or [bõz ˈdʒiɐs] ('good days').
• If the next word begins with a vowel, the final sibilant is treated as intervocalic, and pronounced [z]; bons amigos [bõz ɐˈmiɣuʃ] or [bõz aˈmiɡus] ('good friends').

When two identical sibilants appear in sequence within a word, they reduce to a single consonant. For example, nascer, desço, excesso, exsudar are pronounced with [s] by speakers who use alveolar sibilants at the end of syllables, and disjuntor is pronounced with [ʒ] by speakers who use postalveolars. But if the two sibilants are different they may be pronounced separately, depending on the dialect. Thus, the former speakers will pronounce the last example with [zʒ], whereas the latter speakers will pronounce the first examples with [s] if they are from Brazil or [ʃs] if from Portugal (although in relaxed pronunciation one of the siblants may be dropped). This applies also to words that are pronounced together in connected speech:

• sibilant + /s/, e.g., {{lang|pt|as sopas: either [s] (most of Brazil); [ʃs] (Portugal, standard)
• sibilant + /z/, e.g., as zonas: either [z] (mostly in Brazil); [ʒz] (Portugal, standard)
• sibilant + /ʃ/, e.g., as chaves: always [ʃ];
• sibilant + /ʒ/, e.g., os genes: always [ʒ].

Normally, only the three vowels /ɐ/, /i/ (in BP) or /ɨ/ (in EP), and /u/ occur in unstressed final position. If the next word begins with a similar vowel, they merge with it in connected speech, producing a single vowel, possibly long (crasis). Here, "similar" means that nasalization can be disregarded, and that the two central vowels /a, ɐ/ can be identified with each other. Thus,

• /aa, aɐ, ɐa, ɐɐ/ → [a(ː)] (henceforth transcribed [a (a)]); toda a noite [ˈtoða (a)ˈnojtɨ] or [ˈtoda (a) ˈnojtʃi] ('all night'), nessa altura [ˈnɛs awˈtuɾɐ] or [ˈnɛs alˈtuɾɐ] ('at that point').
• /aɐ̃, ɐɐ̃/ → [ã(ː)]) (henceforth transcribed [ã (ã)]); a antiga ('the ancient one') and à antiga ('in the ancient way'), both pronounced [ã (ã)ˈtʃiɡɐ] or [ã (ã)ˈtiɣɐ]. The open nasalized [ã] appears only in this environment.
• /ii, iĩ/ → [i(ː), ĩ(ː)] (henceforth transcribed [i (i), ĩ (ĩ)]); de idade [dʒi (i)ˈdadʒi] or [di (i)ˈðaðɨ] ('aged').
• /ɨɨ/ → [ɨ]; fila de espera [ˈfilɐ ðɨʃˈpɛɾɐ] ('waiting line') (EP only).
• /uu, uũ/ → [u(ː), ũ(ː)] (henceforth transcribed [u (u), ũ (ũ)]); todo o dia [ˈtodu (u)ˈdʒiɐ] or [ˈtoðu (u) ˈðiɐ] ('all day').

If the next word begins with a dissimilar vowel, then /i/ and /u/ become approximants in Brazilian Portuguese (synaeresis):

• /i/ + V → [jV]; durante o curso [duˈɾɐ̃tʃju ˈkuɾsu] ('during the course'), mais que um [majs kjũ] ('more than one').
• /u/ + V → [wV]; todo este tempo [ˈtoˈdwestʃi ˈtẽpu] ('all this time') do objeto [dwobiˈʒɛtu] ('of the object').

In careful speech and in with certain function words, or in some phrase stress conditions (see Mateus and d'Andrade, for details), European Portuguese has a similar process:

• /ɨ/ + V → [jV]; se a vires [sjɐ ˈviɾɨʃ] ('if you see her'), mais que um [majʃ kjũ] ('more than one').
• /u/ + V → [wV]; todo este tempo [ˈtoˈðweʃtɨ ˈtẽpu] ('all this time'), do objeto [dwɔbˈʒɛtu] ('of the object').

But in other prosodic conditions, and in relaxed pronunciation, EP simply drops final unstressed /ɨ/ and /u/ (elision), though this is subject to significant dialectal variation:

• durante o curso [duˈɾɐ̃tu ˈkuɾsu] ('during the course'), este inquilino [ˈeʃtĩkɨˈlinu] ('this tenant').
• todo este tempo [toˈðeʃtɨ ˈtẽpu] ('all this time'), disto há muito [diʃta ˈmũjtu] ('there's a lot of this').

Aside from historical set contractions formed by prepositions plus determiners or pronouns, like à/dà, ao/do, nesse, dele, etc., on one hand and combined cliticpronouns such as mo/ma/mos/mas (it/him/her/them to/for me), and so on, on the other, Portuguese spelling does not reflect vowel sandhi. In poetry, however, an apostrophe may be used to show elision such as in d'água.

In English phonology, sandhi can be seen when one word ends with a vowel, and the next begins with a vowel. An approximant is inserted between them based on the vowel ending the first word: if it is rounded, e.g. [ʊ], a [w] (voiced labial-velar approximant) is inserted. The vowels [iː], [ɪ], and [ɪː] (including [ɛɪ], [ɑɪ], and [ɔɪ]) take a sandhi of [j] (voiced palatal approximant). All other vowels take [ɹ] (voiced alveolar approximant) (see linking and intrusive R). For example, "two eggs" is pronounced [tuːw.ɛɡz], "three eggs" is [θɹiːj.ɛɡz], and "four eggs" is [fɔːɹ.ɛɡz].

In some situations, especially when a vowel is reduced to a schwa, certain dialects may instead use a glottal stop [ʔ]. For example, "gonna eat" may be pronounced as [ɡʌn.əw.iːt], reflecting the [uː] sound that has been reduced, or as [ɡʌn.əɹ.iːt], reflecting the schwa sound, which takes a sandhi of [ɹ], or as [ɡʌn.ə.ʔiːt], using a glottal stop to separate the words. Note that in this case the glottal stop occurs at the start of "eat" rather than at the end of "gonna". A glottal stop sandhi is especially done when wishing to avoid other, more noticeable, sandhi due to stress; if, in the above example, either the last syllable of "gonna" was stressed, or there was particular stress on the word "eat", a glottal stop would generally be the preferred sandhi.

French liaison and enchaînement can be considered forms of external sandhi.^[3] In enchaînement, a word-final consonant, when followed by a word commencing with a vowel, is articulated as though it is part of the following word. For example, sens (sense) is pronounced /sɑ̃s/ and unique (unique) is pronounced /y nik/; sens unique (one-way, as a street) is pronounced /sɑ̃‿sy nik/.

Liaison is a similar phenomenon, applicable to words ending in a consonant that was historically pronounced but that, in Modern French, is normally silent when occurring at the end of a phrase or before another consonant. In some circumstances, when the following word commences with a vowel, the consonant may be pronounced, and in that case is articulated as if part of the next word. For example, deux frères (two brothers) is pronounced /dø fʁɛʁ/ with a silent ⟨x⟩, and quatre hommes (four men) is pronounced /katʁ ɔm/, but deux hommes (two men) is pronounced /dø‿zɔm/.

In Japanese phonology, sandhi is primarily exhibited in rendaku (consonant mutation from unvoiced to voiced when not word-initial, in some contexts) and conversion of つ or く (tsu, ku) to a geminate consonant (orthographically, the sokuon っ), both of which are reflected in spelling – indeed, the っ symbol for gemination is morphosyntactically derived from つ, and voicing is indicated by adding two dots as in か／が ka, ga, making the relation clear. It also occurs much less often in renjō (連声), where, most commonly, a terminal /n/ on one morpheme results in an /n/ (or /m/) being added to the start of the next morpheme, as in 天皇: てん ＋ おう → てんのう (ten + ō = tennō), meaning "emperor"; that is also shown in the spelling (the kanji do not change, but the kana, which specify pronunciation, change).

Korean has sandhi which occurs in the final consonant or consonant cluster, such that a morpheme can have two pronunciations depending on whether or not it is followed by a vowel. For example, the root 읽 /ik/, meaning ‘read’, is pronounced /ik/ before a consonant, as in 읽다 /ik.ta/, but is pronounced like /il.k/ before vowels, as in 읽으세요 /il.kɯ.se̞.jo/, meaning ‘please read’. Some roots can also aspirate following consonants, denoted by the letter ㅎ (hieut) in the final consonant. This causes 다 /tɐ/ to become /tʰɐ/ in 않다 /ɐntʰɐ/, ‘to not be’.^[4]

As Tamil is strongly characterised by diglossia: there are two separate registers varying by socioeconomic status, a high register and a low one.^[5][6] This in turn adds an extra layer of complexity forming Sandhi.^[7] Tamil employs Sandhi for certain morphological and syntactic structures.^[1][2]

The vowel sandhi occurs when words or morphemes ending in certain vowels are followed by morphemes beginning with certain vowels. Consonant glides (Tamil: ய், romanized: Y and Tamil: வ், romanized: V) are then inserted between the vowels in order to 'smooth the transition' from one vowel to another.^[7]

"The choice of whether the glide inserted will be (ய், Y and வ், V) in Tamil is determined by whether the vowel preceding the glide is a front vowel such as Tamil: இ, ஈ, எ, ஏ or ஐ, romanized: i, ī, e, ē or ai or a back vowel, such as Tamil: உ, ஊ, ஒ, ஓ, அ or ஆ, romanized: u, ū, o, ō, a or ā."^[7]

^[7]

In rapid speech, especially in polysyllabic words: Tamil: இந்த்யாவுலேருந்து, romanized: Intyāvulēruntu, lit. 'From India' may become — இந்த்யாலெருந்து, Intyāleruntu, which may then be further simplified to இந்த்யாலெந்து, Intyālentu.^[7]

In lateral-stop clusters, the lateral assimilates to the stop's manner of articulation, before c, ṇ too becomes ṭ, eg. nal-mai, kal-kaḷ, vaṟaḷ-ci, kāṇ-ci, eḷ-ney > naṉmai, kaṟkaḷ, vaṟaṭci, kāṭci, eṇṇey (ṟ was historically a plosive).

In Spoken Tamil the final laterals, nasals or other sonorants may lose the final position. The final retroflex laterals for pronouns and their PNG markers for example Tamil: ள், romanized: ḷ of (female gender marker) are deleted: (To indicate the omitted stop-consonant is covered in parantheses): Tamil: அவ(ள்) போறா(ள்), romanized: Ava(ḷ) pōṟā(ḷ), lit. 'She goes'.^[7]

In some nouns, sandhi is triggered by the addition of a case ending to the stem.^[7]

In compounding, if the first word ends with /i, u/ and the second word starts with a vowel, the i, u become glides y, v, eg. su-āgata > svāgata. If a word ends with /a, aː/ and the second word begins with /i, u/ they become /eː, oː/, eg. mahā-utsava > mahotsava; if the latter vowel is long, it becomes /ai, au/, eg. pra-ūḍha > prauḍha.

The visarga becomes a /r/ before voiced phones, eg. duḥ-labha > durlabha. Anusvara + plosive makes it a homorganic nasal, before a fricative or /r/ it nasalizes the previous vowel and before /j, ʋ/ it nasalizes the /j, ʋ/.