The device mentioned earlier that produces "mechanical speech" would contain all the words of English, each having been recorded in isolation. A significant difference in natural connected speech is the way that sounds belonging to one word can cause changes in sounds belonging to neighboring words. Assuming that we know how the phonemes of a particular word would be realized when the word is pronounced in isolation, in cases where we find a phoneme realized differently as a result of being near some other phoneme belonging to a neighboring word we call this difference an instance of assimilation. Assimilation is something which varies in extent according to speaking rate and style: it is more likely to be found in rapid, casual speech and less likely in slow, careful speech. Sometimes the difference caused by assimilation is very noticeable, and sometimes it is very slight. Generally speaking, the cases that have most often been described are assimilations affecting consonants. As an example, consider a case where two words are combined, the first of which ends with a single final consonant (which we will call C^f) and the second of which starts with a single initial consonant (which we will call Cⁱ); we can construct a diagram like this:

- - - - - - Cf | Cⁱ - - - - - -

               Word

            boundary

If C^f changes to become like Cⁱ in some way, then the assimilation is called regressive (the phoneme that comes first is affected by the one that comes after it); if Cⁱ changes to become like C^f in some way, then the assimilation is called progressive. An example of the latter is what is sometimes called coalescence, or coalescent assimilation: a final t, d and an initial j following often combine to form ʧ, ʤ, so that 'not yet' is pronounced nɒtʃet and 'could you' is kʊdƷu. In what ways can a consonant change? We have seen that the main differences between consonants are of three types:

i) differences in place of articulation;

ii) differences in manner of articulation;

iii) differences in voicing.

In parallel with this, we can identify assimilation of place, of manner and of voicing in consonants. Assimilation of place is most clearly observable in some cases where a final consonant (C^f) with alveolar place of articulation is followed by an initial consonant (Cⁱ) with a place of articulation that is not alveolar. For example, the final consonant in 'that' ðæt is alveolar t. In rapid, casual speech the t will become p before a bilabial consonant, as in: 'that person' ðæp pз:sṇ; 'that man' ðæp mæn; 'meat pie' mi:p paɪ. Before a dental consonant, t will change to a dental plosive, for which the phonetic symbol is t, as in: 'that thing' ðæt ̪  θɪŋ; 'get those' get ̪ ðʊəz; 'cut through' kΛt ̪ θru:. Before a velar consonant, the t will become k, as in: 'that case' ðæk keIs, 'bright color' braɪk kΛlə, 'quite good' kwaɪk gʊd. In similar contexts d would become b, d and g, respectively, and n would become m, n and ŋ; examples of this would be: 'good boy' gʊb bɔɪ, 'bad thing' bæd ̪ θɪŋ, 'card game' ka:g geɪm, 'green paper' gri:m peɪpə, 'fine thought' faɪn ̪ θɔ:t, 'ten girls' teg gз:lz. However, the same is not true of the other alveolar consonants: s and z behave differently, the only noticeable change being that s becomes S, and z becomes Z when followed by S or j, as in: 'this shoe' ðɪʃ ʃu:; 'those years' ðəʊƷ jɪəz. It is important to note that the consonants that have undergone assimilation have not disappeared; in the above examples, the duration of the consonants remains more or less what one would expect for a two-consonant cluster. Assimilation of place is only noticeable in this regressive assimilation of alveolar consonants; it is not something that foreign learners need to learn to do.

Assimilation of manner is much less noticeable, and is only found in the most rapid and casual speech; generally speaking, the tendency is again for regressive assimilation and the change in manner is most likely to be towards an "easier" consonant - one which makes less obstruction to the airflow. It is thus possible to find cases where a final plosive becomes a fricative or nasal (e.g. 'that side' ðæs saɪd, 'good night' gun naɪt), but most unlikely that a final fricative or nasal would become a plosive. In one particular case we find progressive assimilation of manner, when a word-initial ð follows a plosive or nasal at the end of a preceding word: it is very common to find that the Cⁱ becomes identical in manner to the C^f but with dental place of articulation. For example (the arrow symbol means "becomes"):

'in the'                   in  ðə                       → in ̪n ̪ə

'get them'             get  ðəm                 → get ̪  t ̪əm

'read these'           ri:d ði:z                   → ri:d ̪  d ̪i:z

The ð phoneme frequently occurs with no discernible friction noise

Assimilation of voice is also found, but again only in a limited way. Only regressive assimilation of voice is found across word boundaries, and then only of one type; since this matter is important for foreign learners we will look at it in some detail. If C^f is a lenis (i.e. "voiced") consonant and Cⁱ is fortis ("voiceless") we often find that the lenis consonant has no voicing; for example in 'I have to' the final v becomes voiceless f because of the following voiceless t in aɪ hæf tu, and in the same way the z in 'cheese' ʧi:z becomes more like s when it occurs in 'cheesecake' ʧi:skeɪk. This is not a very noticeable case of assimilation, since, initial and final lenis consonants usually have little or no voicing anyway; these devoiced consonants do not shorten preceding vowels as true fortis consonants do. However, when C^f is fortis ("voiceless") and Cⁱ lenis ("voiced"), a context in which in many languages C^f would become voiced, assimilation of voice never takes place; consider the following example: 'I like that black dog' aI laIk  ðæt blæk dog. It is typical of many foreign learners of English that they allow regressive assimilation of voicing to change the final k of 'like' to g, the final t of 'that' to d and the final k of 'black' to g, giving aI laIg ðæd blæg dɒg. This creates a strong impression of a foreign accent.

Up to this point we have been looking at some fairly clear cases of assimilation across word boundaries. However, similar effects are also observable across morpheme boundaries and to some extent also within the morpheme. Sometimes in the latter case it seems that the assimilation is rather different from the word-boundary examples; for example, if in a syllable-final consonant cluster a nasal consonant precedes a plosive or a fricative in the same morpheme, then the place of articulation of the nasal is always determined by the place of articulation of the other consonant; thus: 'bump' bΛmp, 'tenth' ten̪θ, 'hunt' hΛnt, 'bank' bæŋk. It could be said that this assimilation has become fixed as part of the phonological structure of English syllables, since exceptions are almost non-existent. A similar example of a type of assimilation that has become fixed is the progressive assimilation of voice with the suffixes s, z; when a verb carries a third person singular '-s' suffix, or a noun carries an '-s' plural suffix or an '-'s' possessive suffix, that suffix will be pronounced as s if the preceding consonant is fortis ("voiceless") and as z if the preceding consonant is lenis ("voiced"). Thus:

'cats' kæts                      'dogs' dɒgz

'jumps' ʤΛmps             'runs' rΛnz

'Pat's' pæts                    'Pam's' pæmz

Assimilation creates something of a problem for phoneme theory: when, for example, d in 'good' gʊd becomes g in the context 'good girl', giving gʊg gз:l or b in the context 'good boy' gʊb bɔi, should we say that one phoneme has been substituted for another? If we do this, how do we describe the assimilation in 'good thing', where d becomes dental d̪ before the θ of 'thing', or in 'good food', where d becomes a labiodental plosive before the f in 'food'? English has no dental or labiodental plosive phonemes, so in these cases, although there is clearly assimilation, there could not be said to be a substitution of one phoneme for another. The alternative is to say that assimilation causes a phoneme to be realized by a different allophone; this would mean that, in the case of gʊg  gз:l and gʊb  bɔɪ, the phoneme d of 'good' has velar and bilabial allophones. Traditionally, phonemes were supposed not to overlap in their allophones, so that the only plosives that could have allophones with bilabial place of articulation were p, b; this restriction is no longer looked on as so important. The traditional view of assimilation as a change from one phoneme to another is, therefore, naive: modern instrumental studies in the broader field of coarticu-Iation show that when assimilation happens one can often see some sort of combination of articulatory gestures. In 'good girl', for example, it is not a simple matter of the first word ending either in d or in g, but rather a matter of the extent to which alveolar and/or velar closures are achieved. There may be an alveolar closure immediately preceding and overlapping with a velar closure; there may be simultaneous alveolar and velar closure, or a velar closure followed by slight contact but not closure in the alveolar region. There are many other possibilities.

Much more could be said about assimilation but, from the point of view of learning or teaching English pronunciation, to do so would not be very useful. It is essentially a natural phenomenon that can be seen in any sort of complex physical activity, and the only important matter is to remember the restriction, specific to English, on voicing assimilation mentioned above.