In French, most written word-final consonants are silent in most contexts. Liaison is the pronunciation of such a consonant immediately before a following vowel sound. For example, the letter s in the word les ("the") is generally silent, but it is pronounced /z/ in the combination les amis ("the friends"). In certain syntactic contexts, liaison is impossible; in others, it is obligatory; and in yet others, it is possible but not obligatory.
Like elision (as in *je aime → j'aime), it can be characterized functionally as a euphonic strategy for avoiding hiatus. If we look at it like this, we are adopting a synchronic approach. This approach does not explain cases where the first word already ends in a consonant, such as tels‿amis, and is therefore already perfectly euphonic.
It is also possible to analyse liaison diachronically. With this approach, the liaison consonant has always been there since the days of Latin, and has merely been elided in other contexts over time. So, the s pronounced in mes amis can be seen as simply preserving the s that was always pronounced in meos amicos. Seen in this way, it is mes frères that is exceptional, having lost the s that was pronounced in meos fratres.
For example, final consonants are pronounced as follows in the case of liaison (the transcription uses IPA; in IPA, liaison is indicated by placing an undertie [‿] between the consonant and the vowel):
Liaisons with [.t‿] and [.z‿] are also found with words ending graphically in -t and -z (e.g. tout "all", venez "come"). Liaisons in [.k‿] with the few words ending in a silent -c are limited to some compound words like porc-épic ("porcupine").
With most words whose spellings end in -n and whose pronunciations end in nasal vowels ([ɑ̃], [ɛ̃], [œ̃], or [ɔ̃]), the vowel will be denasalized during liaison:
However, this liaison in -n (and its eventual denasalisation) should not be done before adverbs like ou, et (whose final -t also never produces any liaison).
Liaison with words ending in -er also frequently (but not systematically) leads to a change in vowel quality:
Some rare words ending with a silent -l like persil will also create a liaison in [.l‿], just like words ending in non silent final [l(ə)] (optionally followed by a mute e).
Finally, the words trop [tʁo] ("too") and beaucoup [bo.ku] ("much") productively allow liaison with the final consonant [.p‿]: un prix trop élevé... ("a too high price"), il a beaucoup à apprendre ("he has much to learn'').
The primary requirement for liaison at a given word boundary is of course the phonological or lexical identity of the words involved: The preceding word must supply a potential liaison consonant and the following word must be vowel-initial (and not exceptionally marked as disallowing liaison; see the discussion of "aspirated h" below). The actual realization of liaison, however, is subject to interacting syntactic, prosodic, and stylistic constraints.
Grammatical descriptions of French identify three kinds of liaison contexts: Those where liaison is obligatory, those where it is impossible, and those where it is optional. Pedagogical grammars naturally emphasize what is obligatory or forbidden, and these two categories tend to be artificially inflated by traditional prescriptive rules. Speakers' natural behavior in spontaneous speech shows that in fact relatively few contexts can be said to systematically give rise to, or fail to give rise to, liaison. Any discussion of liaison must take both descriptive and prescriptive perspectives into account, because this is an area of French grammar where speakers can consciously control their linguistic behavior out of an awareness of how their speech diverges from what is considered "correct".
We can identify a small number of contexts where speakers consistently produce liaison in all speech styles, and where the absence of liaison is immediately perceived as an error of pronunciation. These are the contexts where liaison is truly obligatory:
Note that the first two contexts also require obligatory vowel elision for the relevant determiners and pronouns (le, la, me, se, etc.)
The following contexts are often listed as obligatory liaison contexts, but they are more accurately characterized as contexts where liaison is frequent:
Specific instances of these combinations reveal varying tendencies. For certain lexical items (e.g. petit, très), speakers may have a preference for liaison approaching that of the obligatory liaison contexts.
|uninverted form||inverted form||translation|
|elle dort [ɛldɔʁ]||dort-elle [dɔʁtɛl]||she sleeps|
|il vend [ilvɑ̃]||vend-il [vɑ̃til]||he sells|
|ils parlent [ilpɑʁl]||parlent-ils [pɑʁltil]||they speak|
|on parle [ɔ̃pɑ̃ʁl]||parle-t-on [pɑ̃ʁltɔ̃]||one speaks|
The appearance of this consonant in modern French can be described as a restoration of the Latin 3rd person singular ending -t, under the influence of other French verbs that have always maintained final -t.
The earliest examples of this analogical t in writing date to the mid-15th century, although this practice (and the corresponding pronunciation) was not fully accepted by grammarians until the 17th century (Holbrook 1923).
Grammars mention other contexts where liaison is "forbidden", despite (or precisely because of) the fact that speakers sometimes do produce them spontaneously.
All remaining contexts can be assumed to allow liaison optionally, although exhaustive empirical studies are not yet available. Preferences vary widely for individual examples, for individual speakers, and for different speech styles. The realization of optional liaisons is a signal of formal register, and pedagogical grammars sometimes turn this into a recommendation to produce as many optional liaisons as possible in "careful" speech. The conscious or semi-conscious application of prescriptive rules leads to errors of hypercorrection in formal speech situations (see discussion below).
Conversely, in informal styles, speakers will semi-consciously avoid certain optional liaisons in order not to sound "pedantic" or "stilted". Other liaisons lack this effect. For example Ils ont‿attendu ("they have waited") is less marked than tu as‿attendu ("you have waited"), and neither liaison is likely to be realized in highly informal speech (where one might instead hear [i(l)zɔ̃atɑ̃dy] and [taatɑ̃dy], or simply [taːtɑ̃dy].) On the other hand, the liaison in pas‿encore can be either present or absent in this register.
On the other end, producing a liaison where one is impossible is perceived as an error. For example, pronouncing a liaison consonant instead of respecting hiatus before an aspirated h is taken to indicate an uncultivated or unsophisticated speaker. While all speakers know the rule, they may have incomplete knowledge about which words it must apply to. The effect is less noticeable with rare words (such as hiatus itself), which many speakers may not spontaneously identify as aspirated h words.
Errors due to hypercorrection or euphony are also observed: a liaison is pronounced where it doesn't exist (where it is possible by spelling, but forbidden, as with et‿ainsi, or where it is impossible even by spelling, as with moi-z-avec). This phenomenon is called pataquès. In rare cases, these liaisons may be conserved by the language and become obligatory, such as in donnes-en. Otherwise, they are perceived in the same way as omissions of disjunction, suggesting an "uncultivated" speaker or extremely informal speech. Such an error is sometimes called cuir ("leather") when the inserted consonant is [t], velours ("velvet") when it is [z], although dictionaries do not all agree on these terms:
Careful pronunciation (but without the obligatory reading of “null e’s”) is necessary in a formal setting. The voice is a tool of persuasion: it reflects, through a pronunciation perceived as correct (according to prevailing norms), intellectual qualities, culture, self-control, and wit. Pushed too far, the over-proliferation of liaisons can render a speech ridiculous. It has been pointed out that French politicians and speakers (Jacques Chirac, for example) pronounce some liaison consonants, independently of the following word, introducing a pause or a schwa afterwards. For example, ils ont entendu (“they heard”) is normally pronounced [ilz‿ɔ̃‿ɑ̃tɑ̃dy] or, in more careful speech, [ilz‿ɔ̃t‿ɑ̃tɑ̃dy]. A speaker using this "politician" pronunciation would say (where [ǀ] represents a pause; ils ont'… entendu). One might even hear ils ont décidé (“they decided”) pronounced (ils ont’… décidé) or (ils onteuh… décidé). In the first example, we have liaison without enchaînement, not the normal configuration in ordinary speech. In the second, the liaison is completely non-standard, since it introduces a liaison consonant before another consonant.
However, the ancient final [t] of grand did not cease to be pronounced when the following word began with a vowel and belonged to the same tonic cell; It is effectively not at the end of the word anymore, since the ear identifies the stressed group (formed by univerbation), in which the final consonant and the initial vowel appear together, as a new group (or "word") within which the consonant in question has ceased to be final. Bearing in mind that stress in French falls on the last syllable of a word, or a group of words when they are bound grammatically, this situation can be symbolized as follows (the acute represents stress):
This has to do with what the hearer considers to be a word. If grand homme is analyzed as [gʁɑ̃t‿ɔm], the ear in fact understands [gʁɑ̃'tɔm], a continuous group of phonemes whose tonic accent signals that they form a unit. It is possible to make a division as instead of . Then this [t] will no longer be felt to be a final consonant but a pre-stress intervocalic consonant, and therefore it will resist the deletion that it would undergo if it were at the end of a stressed syllable. It can however undergo other modifications thereafter.
The written form, though, was adapted to criteria that are not phonetic, but etymological (among others): where grand is written, [gʁɑ̃t] is pronounced in front of certain vowels, without that being really awkward: the maintenance of the visual alternation -d ~ -de is more productive.
The other cases are explained in a similar fashion: sang, for example, was pronounced [sɑ̃ŋk] (and written sanc) in Old French, but the final -g has replaced the -c in order to recall the Latin etymology, sanguis, and derivatives like sanguinaire, sanguin. Currently this liaison is almost never heard except in one part of the singing of the Marseillaise ("qu'un san(g) /k/ impur") or in the expression "suer sang et eau". Outside those, the hiatus is tolerated.
Finally, the case of -s and -x pronounced [z] in liaison is explained differently. One must be aware, firstly, that word-final -x is a medieval shorthand for -us (in Old French people wrote chevax for chevaus, latter written chevaux when the idea behind this -x was forgotten). The sound noted -s and -x was a hard [s], which did not remain in French after the twelfth century (it can be found in words like (tu) chantes or doux), but which was protected from complete elision when the following word began with a vowel (which effectively means, when it was found between two vowels). However, in French, such [s] is voiced and becomes [z] (which explains why, in words like rose and mise, the s is pronounced [z] and not [s]).
For example, the Prayer by Gilles Vaudelin (a document compiled in 1713 using a phonetic alphabet, and introduced in the Nouvelle maniere d'écrire comme on parle en France ["A New Way of Writing as We Speak in France"]), probably representative of oral language, maybe rural, of the time, shows the absence of the following liaisons (Vaudelin's phonetic alphabet is transcribed using equivalent IPA):