A First South African Corpus of Multilingual Code-switched Soap Opera Speech

A First South African Corpus of Multilingual Code-switched Soap Opera Speech Ewald van der Westhuizen, Thomas Niesler Department of Electrical and Electronic Engineering Stellenbosch University, South Africa ewaldvdw@sun.ac.za, trn@sun.ac.za Abstract We introduce a speech corpus containing multilingual code-switching compiled from South African soap operas. The corpus contains English, isiZulu, isiXhosa, Setswana and Sesotho speech, paired into four language-balanced subcorpora containing English-isiZulu, English-isiXhosa, English-Setswana and English-Sesotho. In total, the corpus contains 14.3 hours of annotated and segmented speech. The soap opera speech is typically fast, spontaneous and may express emotion, with a speech rate that is between 1.22 and 1.83 times higher than prompted speech in the same languages. Among the 10343 code-switched utterances in the corpus, 19207 intrasentential language switches are observed. Insertional code-switching with English words is observed to be most frequent. Intraword code-switching, where English words are supplemented with Bantu affixes in an effort to conform to Bantu phonology, is also observed. Most bigrams containing code-switching occur only once, making up between 64% and 92% of such bigrams in each subcorpus. Keywords: code-switching, spontaneous speech, South African languages, isiZulu, isiXhosa, Setswana, Sesotho 1. Introduction South Africa has 11 official languages and since the ma- jority of South Africans are multilingual, code-switching occurs commonly and spontaneously. Code-switching is the phenomenon of using more than one language within the same conversation or utterance (Van Dulm, 2007). Lan- guage switches may even occur mid-word, as examples in our data will demonstrate. In an effort to be representative and accommodate a large viewer base, conversations in South African soap operas are multilingual and exhibit code-switching. Soap operas are therefore an interesting source of multilingual code- switched speech data. The corpus we describe in this pa- per displays various types of code-switching. Intersenten- tial code-switching occurs when the language alternates be- tween utterances of a conversation. For example, a speaker may utter a sentence in English while the following sen- tence is spoken in isiZulu. Intrasentential code-switching occurs when the language alternates within a single ut- terance. For example, an utterance may start in English and mid-sentence switch to isiZulu. Intrasentential code- switching can further be divided into alternational and in- sertional switching. Alternational code-switching occurs when an utterance starts with a phrase in one language and switches to a phrase in another language and where the phrases conform to the grammar of the respective lan- guages. Insertional code-switching occurs when a language element from a secondary language is embedded into the structure of a matrix language. Code-switching can also oc- cur at morpheme-level within word boundaries (intraword). Examples of intraword switching from our corpus include the joining of Bantu language class affixes to English stems. Currently our corpus contains 14.3 hours of language- balanced speech compiled from soap opera broad- casts. Four Bantu languages are paired with English to yield subcorpora containing English-isiZulu, English- isiXhosa, English-Setswana and English-Sesotho code- switched speech. IsiZulu and isiXhosa are part of the Nguni language family, while Setswana and Sesotho are part of the Sotho-Tswana language family. …

Content truncated to 500 words