Orality and Difficulties in the Transcription of Spoken Corpora (2)

Description of the corpus

C-ORAL-ROM is a multilingual spontaneous speech corpus (Cresti et al., 2002) of the four main roman languages: French, Italian, Portuguese and Spanish. Each subcorpus consists of around 300,000 words. With the aim of enabling comparability between the different subcorpora, several sampling criteria concerning the distribution of the corpus were established: as long as each of the variation parameters is fully present in the corpus, the linguistic variation will be well represented (Moreno, 2002). In that sense, two elements are to be considered as basic elements: on one hand, the characteristics of the speakers and, on the other hand, the context of use. As far as the speakers are concerned, age, sex, education, occupation and geographical origin were taken into account. As for the contexts of use, a basic distinction was made between the dialogic structure (monologues and dialogues or conversations) and kind of situation (familiar or public).

A second important distinction was made between formal and informal speech. Each is represented in the corpus by 50% of the texts. Inside the informal part, a distinction was made between the familiar and the public domains: the first is represented by 75% of the texts, while the public domain accounts for the other 25%. As for the formal speech, the distribution of the texts was made following a thematic criterion: the natural context formal speech area (43% of the texts) is formed by recordings such as conferences, political debates, political speeches, sermons, professional explanations and texts dealing with business, law and teaching. In the same way, the texts which are part of the formal speech in media section (40% of the texts) are grouped in the following categories: interviews, meteo, news, reportages, scientific press, sports and talk-shows. Finally, inside the formal speech part, a section made up of phone recordings (17 % of the texts) is included.

Other relevant criteria concerning the corpus design are: acoustic quality of the samples (all are digital recordings), legal status (recording, transcription and publishing were done after the written authorization of all participants) and spontaneity of the recordings (no previous scripts were used and there were no restrictions in the use of the language and the expression of opinions).