Tone sandhi is a phonological change occurring in tonal languages, in which the tones assigned to individual words or morphemes change based on the pronunciation of adjacent words or morphemes. It usually simplifies a bidirectional tone into a one-direction tone. It is a type of sandhi, or fusional change, from the Sanskrit word for "joining".
Tone sandhi occurs to some extent in nearly all tonal languages, manifesting itself in different ways. Tonal languages, characterized by their use of pitch to affect meaning, appear all over the world, especially in the Niger-Congo language family of Africa, and the Sino-Tibetan language family of East Asia, as well as other East Asian languages such as Tai-Kadai, Vietnamese, and Papuan languages. Tonal languages are also found in many Oto-Manguean and other languages of Central America, as well as in parts of North America (such as Athabaskan in British Columbia, Canada), and Europe.
Many North American and African tonal languages undergo "syntagmatic displacement," as one tone is replaced by another in the event that the new tone is present elsewhere in the adjacent tones. Usually, these processes of assimilation occur from left to right. In the Bantu languages of West Africa, for example, an unaccented syllable takes the tone from the closest tone to its left. However, in East and Southeast Asia, "paradigmatic replacement" is a more common form of tone sandhi, as one tone changes to another in a certain environment, whether or not the new tone is already present in the surrounding words or morphemes.
Many languages spoken in China have tone sandhi; some of it quite complex.Southern Min (Minnan), which includes Hokkien, Taiwanese, and Teochew, has a complex system, with in most cases every syllable changing into a different tone, and which tone it turns into depending on the final consonant of the syllable that bears it.
Take for example Taiwanese varieties of Hokkien, which have two tonemes (numbered 4 and 8 in the diagram above) that occur in checked syllables (those ending in a stop consonant) and five tonemes in syllables that do not end in a stop. In Taiwanese, within a phonological phrase, all its non-neutral-tone syllables save for the last undergo tone sandhi. Among the unchecked syllables, tone 1 becomes 7, 7 becomes 3, 3 becomes 2, and 2 becomes 1. Tone 5 becomes 7 or 3, depending on dialect. Stopped syllables ending in Pe?h-?e-j? ⟨p⟩, ⟨t⟩, or ⟨k⟩ take the opposite tone (phonetically, a high tone becomes low, and a low tone becomes high) whereas syllables ending in a glottal stop (written in Pe?h-?e-j? by ⟨h⟩) drop their final consonant to become tone 2 or 3.
The seven or eight tones of Hmong demonstrate several instances of tone sandhi. In fact the contested distinction between the seventh and eighth tones surrounds the very issue of tone sandhi (between glottal stop (-m) and low rising (-d) tones). High and high-falling tones (marked by -b and -j in the RPA orthography, respectively) trigger sandhi in subsequent words bearing particular tones. A frequent example can be found in the combination for numbering objects (ordinal number + classifier + noun): ib (one) + tus (classifier) + dev (dog) > ib tug dev (note tone change on the classifier from -s to -g).
Tone sandhi is compulsory as long as the environmental conditions that trigger it are met. It is not to be confused with tone changes that are due to derivational or inflectional morphology. For example, in Cantonese, the word "sugar" (?) is pronounced tòhng (/t:/ or /t:/, with low (falling) tone), whereas the derived word "candy" (also written ?) is pronounced tóng (/t:/, with mid rising tone). Such a change is not triggered by the phonological environment of the tone, and therefore is not an example of sandhi. Changes of morphemes in Mandarin into its neutral-tone are also not examples of tone sandhi.
In Hokkien (again exemplified by Taiwanese varieties), the words kia? (whose base-tone is high flat and means 'to be afraid') plus lâng (whose base-tone is upwardly curving and means 'person, people') can combine via either of two different tonal treatments with a corresponding difference in resulting meaning. A speaker can pronounce the lâng as a neutral-tone (= low here), which causes the kia? (because it does not go into neutral-tone and immediately precedes) to keep its base-tone (which differs, by the way, from for instance the Standard-Chinese third tone preceding a neutral-tone's more common of its two behaviors). The result means 'to be frightful' (written in Pe?h-?e-j? as kia?--lâng). However, one can instead NOT make the lâng neutral-tone, which both causes it (due to being the final non-neutral tone in the phonological phrase) to keep its base-tone and causes the kia? (because this comes earlier in the phrase than the last non-neutral-tone syllable) to take its sandhi-tone (that being mid flat, because its base-tone is high-flat). This latter treatment means 'to be frightfully dirty' or 'to be filthy' (written as kia?-lâng). The former treatment does not involve the application of the tone sandhi rules, but rather is a derivational process (that is, it derives a new word), while the latter treatment does involve the tone sandhi, which applies automatically (here as anywhere that a phrase contains more than one syllable not made neutral-tone).
Standard Chinese (Standard Mandarin) features several tone sandhi rules:
Yatzachi Zapotec, an Oto-Manguean language spoken in Mexico, has three tones: high, middle and low. The three tones, along with separate classifications for morpheme categories, contribute to a somewhat more complicated system of tone sandhi than in Mandarin. There are two main rules for Yatzachi Zapotec tone sandhi, which apply in this order:
Molinos Mixtec, another Oto-Manguean language, has a much more complicated system of tone sandhi. The language has three tones (high, mid, or low, or 1, 2, or 3, respectively), and all roots are disyllabic, meaning that there are nine possible combinations of tones for a root or "couplet". The tone combinations are expressed here as a two digit number (high-low is represented as 13). Couplets are also classified into either class A or B, as well as verb or non-verb. A number of specific rules depending on these three factors determine tone change. One example of a rule follows:
"Basic 31 becomes 11 when following any couplet of Class B but does not change after Class A (except that after 32(B') it optionally remains 31"
?a²?a² (class B) "chiles" + ?i³?i¹ (class A) "dry" > ?a²?a²?i¹?i¹ "dry chiles"
Akan, a Niger-Congo language spoken in Ghana, has two tones: high and low. The low tone is default. In Akan, tones at morpheme boundaries assimilate to each other through tone sandhi, the first tone of the second morpheme changing to match the final tone of the first morpheme.
Sinologists sometimes use reversed Chao tone letters to indicate sandhi, with the left-facing letters of the IPA on the left for the underlying tone, and reversed right-facing letters on the right for the realized tone. For example, the Mandarin example of n? [ni] + h?o [xa?] > ní h?o [nixa?] above would be transcribed:
The individual Unicode components of the reversed tone letters are ⟨? ? ? ? ?⟩.