Bulu_ALCAM-MultimodalDataset

Language

Bulu (also written Boulou) is a Bantu language spoken in southern Cameroon. The Administrative Atlas of Cameroon's Languages (Breton and Bikia Fohtung, 1991) classifies Bulu as part of the Beti-Fang dialect cluster. However, the Bulu-speaking community generally contests this classification, asserting that Bulu constitutes a distinct language rather than a mere dialect of an overarching Beti-Fang macro-linguistic entity. This position is reinforced by the existence of a well-established, autonomous literary tradition in Bulu dating back to at least the early 20th century, most notably represented by Jean Louis Njemba Medu's Nnanga Kon (1939), a landmark of Bulu prose literature.

Bulu is spoken primarily in four administrative departments of Cameroon's South Region: the Ntem Division, the Mvila Division, the Dja-and-Lobo Division, and the Océan Division. Speakers are found in the major towns of these divisions, including Ebolowa (headquarters of the Mvila Division), Sangmelima (headquarters of the Dja-and-Lobo Division), Ambam, and Kribi.

Variants

The Bulu-speaking community generally identifies with two broad speech areas, sometimes described as dialects:

• The Ebolowa speech area, centred on Ebolowa, the administrative and commercial hub of the Mvila Division. This variety is widely regarded as a reference form of Bulu and has been the basis for much of the written production and missionary linguistic work on the language.

• The Sangmelima speech area, centred on Sangmelima, headquarters of the Dja-and-Lobo Division. This variety shares the core phonological and grammatical system of the Ebolowa speech area while exhibiting some lexical and phonetic differences.

Both speech areas are mutually intelligible and share the same broad linguistic structure. The variety represented in this dataset was elicited through a structured questionnaire targeting core lexical and grammatical categories, and is consistent with the standard Bulu used in existing written resources.

Writing System

The writing system used for the transcription of Bulu in this dataset is the International Phonetic Alphabet (IPA), as reflected in the lexical entries and sentence-level examples in the datasheet. The phonological inventory described below is derived directly from the attested forms in the dataset.

1. Vowels

The vowel system attested in the dataset is as follows:

i, e, ɛ, a, ɔ, o, u, ə

These vowels occur both in short and long forms, and with distinct tone markings in lexical items and running text (e.g. mə́ndíp 'water', àyàp 'bird', ènə̀ 'it is', ə́lí 'child').

A nasalised vowel is also attested in the dataset, indicated by a subscript tilde: ə̰ (e.g. ə̰̀nkɔ́t, ə̰̀dzálán).

Vowel length is indicated in the dataset by a colon following a vowel symbol (e.g. é:nə̀, ɲɔ̌:ŋ).

The diphthong bwì is also attested (e.g. àbwì).

2. Consonants

The consonant inventory reflected in the dataset includes the following simple, prenasalized, palatal, and affricate consonants:

b, d, dz, f, g, k, kp, l, m, mb, mf, mv, n, nd, ng, ngb, nj, ŋ, p, r, s, t, tʃ, v, w, y, z, ɲ

Palatalisation is marked with a superscript ʸ (e.g. bìtʃʸè).

Additional consonant symbols attested include:

• Prenasalised stops and fricatives: mb, nd, ng, ngb, nj, mf, mv

• Doubly articulated stop: kp

• Affricate: tʃ (and its palatalised variant tʃʸ)

• Palatal nasal: ɲ

• Velar nasal: ŋ

An apostrophe is used in some orthographic representations of Bulu (e.g. in the Protestant and Catholic orthographies) to mark glottal stops or syllable boundaries, as seen in the existing Bulu literary tradition (e.g. Njemba Medu's Nnanga Kon). However, the IPA-based transcription used in this dataset does not use the apostrophe for this purpose.

3. Tone system

The dataset encodes lexical and grammatical contrastive tones, marked directly on vowels and on sonorant consonants. The following tonal categories are attested in the dataset:

• High tone (H): á, é, ɛ́, í, ó, ɔ́, ú, ə́

• Low tone (L): à, è, ɛ̀, ì, ò, ɔ̀, ù, ə̀

• Rising contour tone (LH): ǎ, ě, ǐ, ǒ, ɔ̌, ǔ, ə̌

• Falling contour tone (HL): â, ê, î, ô, ɔ̂, û, ə̂

These tonal contrasts are phonemically significant and grammatically functional in Bulu, distinguishing both lexical items and grammatical categories including noun class agreement.

A note on orthographic conventions: the existing standard orthographies for Bulu — the Protestant orthography and the Catholic orthography (inspired by the model laid out by François Pichon in 1950), as well as the General Alphabet of Cameroon's Languages — do not systematically mark tone. The IPA-based transcription employed in this dataset represents a more phonologically explicit encoding intended for linguistic and speech technology applications.

4. Noun class system

The dataset reflects Bulu's Bantu noun class system, visible in the morphological alternations between singular and plural forms of nouns. These are clearly encoded in the paired entries of the datasheet, for example:

• àyàp (singular) / mə̀yàp (plural) — 'bird(s)'

• ènə̀ (singular prefix class) / mìnɛ́n (plural prefix class) — cf. ànɛ́n / mə̀nɛ́n

• èdɔ́k (singular) / bìdɔ́k (plural) — e.g. 'bone(s)'

• ètùn (singular) / bìtùn (plural)

This morphological patterning is central to the Bantu structure of Bulu and makes the dataset particularly valuable for morphological modelling and noun class analysis.

Source

The dataset was collected through a structured questionnaire designed to elicit the core lexicon and grammatical properties of the Bulu language. This was carried out as part of the Atlas Linguistique du Cameroun (ALCAM) project. The questionnaire targets basic vocabulary domains (body parts, animals, natural elements, actions, spatial and temporal notions) and elicits both singular and plural forms of nouns, as well as example sentences illustrating usage in context. The audio recordings were made by a native Bulu speaker.

Domain

The dataset represents a linguistic questionnaire designed to elicit the basic lexicon and grammatical information of the Bulu language. The content reflects elicited speech rather than spontaneous or naturalistic discourse, and covers core vocabulary domains including body parts, animals, natural elements, objects, actions, and basic grammatical constructions. The dataset also includes paired singular–plural noun forms, which are central to the morphological structure of Bulu as a Bantu language.

Size

The dataset comprises three components:

1. Datasheet (ALCAM_dataset_Bulu.tsv): 389 rows across 19 columns, covering 221 unique lexical entries. Of these, 271 rows contain example sentences in Bulu with French equivalents, and 260 rows include morpheme-level glosses. Parts of speech represented include nouns (n), verbs (v), adjectives (adj), numerals (num), pronouns (prn), prepositions (prep), and adverbs (adv).

2. Audio recordings: 384 audio clips in MP3 format, distributed across 2 folders:

• Folder 1 (Bulu_tts_dataset_384clips_2379s_20260406-1522_part1of2): 200 audio files — 11 minutes 18 seconds

• Folder 2 (Bulu_tts_dataset_384clips_2379s_20260406-1522_part2of2): 184 audio files — 12 minutes 29 seconds

GRAND TOTAL: 384 audio files — 23 minutes 47 seconds

Individual audio clips are short, typically ranging from 1 to 9 seconds in duration, consistent with the elicited word- and sentence-level prompts of the questionnaire.

3. Audio–text mapping file: a tab-separated file with 183 lines and 4 columns, linking each audio file to its Bulu transcription.

A detailed breakdown of durations per file is provided in the accompanying duration report.

Structure

The dataset comprises three components:

1. Datasheet (ALCAM_dataset_Bulu.tsv): a tab-separated file with 389 rows and 19 columns. Each row corresponds to a lexical entry (singular or plural form) or an associated example sentence. The columns are:

• OrigID: original number of the lexical entry in the paper questionnaire

• EditID: any modification of OrigID

• FrenchRef: the reference entry as originally provided in French

• FrenchComm: original researcher comments about the reference entry

• French: the lexical entry in French

• Note: researcher notes on the lexical entry

• POS: part of speech (n = noun, v = verb, adj = adjective, num = numeral, prn = pronoun, prep = preposition, adv = adverb)

• Class: noun class (where applicable)

• Morf: morphological attribute (sg = singular, pl = plural)

• Var: variant information (where applicable)

• Word: the lexical entry in Bulu (IPA transcription)

• CrossRef: cross-reference to related lexical entry numbers

• FrenchEx: example sentence in French

• LangEx: example sentence in Bulu (IPA transcription)

• LangExEdit: manually edited version of LangEx

• LangPars: word-for-word morpheme parsing in Bulu

• LangParsEdit: edited version of LangPars

• FrenchPars: French equivalent of LangParsEdit (word-for-word gloss)

• FrenchParsEdit: edited version of FrenchPars

2. Audio recordings: MP3 files, distributed across 2 folders (see Size section above).

3. Audio–text mapping file: a tab-separated file with 183 lines and 4 columns, following the format:

audio_filename.mp3 key sentence attempts

where:

• audio_filename: the name of the MP3 audio file

• key: a unique identifier (matching the filename without extension)

• sentence: the Bulu transcription in IPA, which may contain a lexical entry alone (e.g. àyàp), or a lexical entry followed by a separator and an example sentence (e.g. àyàp ; àyàp élé ánə̀ ə́lí)

• attempts: the number of recording attempts made for that entry

The dataset is designed for linguistic analysis and NLP/speech technology pipelines requiring paired audio-text data.

Sample

Datasheet sample (selected rows from ALCAM_dataset_Bulu.tsv):

Audio–mapping sample (selected rows from the mapping file):