Just as in Birdsong’s (2003) experiment, both words (W) and phrases (P) were used to avoid
bias towards either group – one group worked on prosody, i.e. especially at the sentencelevel,
whereas the other (segmental) group mainly worked at the phoneme-level and wordlevel.
For that matter, it is crucial to use both types of item because this is a pilot experiment.
Therefore, what served as stimuli in the experiment was a set of twenty English words and
sentences – ten words and ten sentences. Appendix C contains a list of all twenty items; the
words are transcribed phonetically, with stress markings, and the phrases are transcribed
phonemically, with stress and nucleus markings.
Both words and phrases illustrated difficulties that French learners typically encounter
with English pronunciation. Following the theoretical account of recurrent production errors
by French EFL learners in Chapter 1, major problematic aspects of English phonology were
selected for the creation of the stimuli. Roughly 50% of the problems were segmental, and
50% were related to prosody. The ten words were either monosyllabic (six of ten), or
disyllabic (four of ten). The ten phrases consisted of eight or nine syllables, and some of them
were taken or adapted from examples in Wells (2006). Among the segmental problems, the
following appeared: the realization of dental fricatives /?/-/ð/ (e.g. W09 “either”, P04
“other”); the lax/tense vowel distinction (e.g. P02 “live” vs. P10 “leaving”); the velar nasal /?/
(e.g. W08 “thinking”). Among the prosodic problems: unexpected nucleus placement;
deaccenting of function words; word stress (e.g. W10 “hel’lo”, P06 “’interesting”). The
phrases displayed auxiliaries in both full and reduced forms, which was designed to observe
if the subjects of the prosodic group could reduce them by themselves. The tables in
Appendix C give a more detailed account of the type of segmental and suprasegmental
difficulties of the stimuli, and the number of syllables in each item. To choose the words, a
balance in their use was kept, so their frequency per million words in the spoken part of the
British National Corpus (BNC) is also given. It should be noted that although it is the
segmental problems of each word and the prosodic problems of each phrase that are listed in
the tables, the phrases obviously contain segmental difficulties that reflect those in the words
(e.g.
difficulties (e.g. the stress pattern of W10 “hel’lo”).
For lack of time, extemporaneous production and perception capacities could not be
tested. As a pilot experiment, read speech only was evaluated through the recording of the
twenty items by the ten French learners of English, and Bertran’s (1999: 109) comment brings
support to this choice: “we believe that a laboratory corpus, made up of several “artificial”
utterances created ad hoc is more reliable, since it permits the isolation of the variables under
study as well as the neutralisation of other factors”. In this respect, no figures were used, and
grammatical or lexical mistakes and hesitations were avoided, all the more as this might have
had an impact on the listeners’ judgements.