In phonology, particularly within historical linguistics, dissimilation is a phenomenon whereby similar consonants or vowels in a word become less similar. In English, dissimilation is particularly common with liquid consonants such as /r/ and /l/ when they occur in a sequence.
When a /r/ sound occurs before another in the middle of a word in rhotic dialects of English, the first tends to drop out, as in "beserk" for berserk, "suprise" for surprise, "paticular" for particular, and "govenor" for governor – this does not affect the pronunciation of government, which has only one /r/, but English government tends to be pronounced "goverment", dropping out the first n.
In English, r-deletion occurs when a syllable is unstressed and /r/ may drop out altogether, as in "deteriate" for deteriorate and "tempature" for temperature, a process called haplology. When the /r/ is found in /bru/, it may change to /j/: "Febyuary" for February, "defibyulator" for defibrillator, though this may be due to analogy with similar words, such as January and calculator (compare nucular, which may have arisen through an analogous process).
An example where a relatively old case of phonetic dissimilation has been artificially undone in the spelling is English colonel, whose standard pronunciation is /ˈkɝnəl/ (with the r sound) in North-American English, or /ˈkɜːnəl/ in RP. It was formerly spelt coronel and is a borrowing from French coronnel, which arose as a result of dissimilation from Italian colonnello.
There are several hypotheses on the cause of dissimilation. According to John Ohala, listeners are confused by sounds with long-distance acoustic effects. In the case of English /r/, rhoticization spreads across much of the word: in rapid speech, many of the vowels may sound as if they had an r. It may be difficult to tell whether a word has one source of rhoticity or two. When there are two, a listener might wrongly interpret one as an acoustic effect of the other, and so mentally filter it out.
This factoring out of coarticulatory effects has been experimentally replicated. For example, Greek pakhu- (παχυ-) "thick" derives from an earlier *phakhu-. When test subjects are asked to say the *phakhu- form in casual speech, the aspiration from both consonants pervades both syllables, making the vowels breathy. Listeners hear a single effect, breathy voiced vowels, and attribute it to one rather than both of the consonants, as they assume the breathiness on the other syllable to be a long-distance coarticulatory effect, thus replicating the historical change in the Greek word.
Dissimilation, like assimilation, may involve a change in pronunciation relative to a segment that is adjacent to the affected segment or at a distance, and may involve a change relative to a preceding or a following segment. As with assimilation, anticipatory dissimilation is much more common than lag dissimilation, but unlike assimilation, most dissimilation is triggered by non-contiguous segments. Also, while many kinds of assimilation have the character of a sound law, few dissimilations do; most are in the nature of accidents that befall a particular lexical item.
When, through sound change, elements of a grammatical paradigm start to conflate in a way that is not easily remedied through re-wording, the forms may dissimilate. For example, in modern Korean the vowels /e/ and /ɛ/ are merging for many people in the capital Seoul, and concurrently the second-person pronoun 네 /ne/ 'you' is shifting to 니 /ni/ to avoid confusion with the first-person pronoun 내 /nɛ/ 'me'.
Similarly, it appears that English she, historically heo, may have acquired its modern sh form through dissimilation from he, though it is not clear whether the mechanism was idiosyncratic sound change (palatalization) of heo, or substitution of heo with the feminine demonstrative pronoun seo.