Chapter 2. French liaison and the lexical repository 49
510(^01125425725102513251625) 1,925
5
10
(^011254257251025) 1,325 1,625 1,925
LOW
5
10
(^011253257251025) 1,325 1,625 1,925
INTERMEDIATE
5
10
(^011254257251025) 1,325 1,625 1,925
HIGH
OCCURRENCES
RANK
Figure 8. Magnification of the tail for the low, intermediate and high education levels.
Taken together, the analyses by age and educational level clearly indicate that
French liaison features a statistical tendency toward a power-laws distribution
independently of the social characteristics of the speakers. This finding sub-
stantially agrees with the results of the analysis by consonant class (see above,
§3.2) inasmuch as the power-laws distribution of enacted liaisons turns out to
resist any statistical manipulation of the investigated corpus. On the other hand,
splitting the data by age or educational level has revealed that different groups
of speakers produce different examples of low frequency liaison environments.
Our analysis does not reveal anything about whether this variability is correlated
with particular social factors. Studies specifically devoted to illustrating the fine-
grained patterns of variation according to these and other social factors will be
more revealing in this respect. However, our analysis does reveal that if we want
to analyse the nature and extent of the variability associated with the production
of liaison, we should preferably look at the tail of the distribution, rather than
at its head (or its body). The variegated sample of low- and very-low-frequency
items is the most likely repository of lexical environments differentially selected
by different groups of speakers.
The size of the corpus over which the analysis is conducted is therefore crucial
in this respect. Our PFC sample has allowed us to uncover the different behaviour
of the head vs. the tail of the curve because it includes such a large number of