KIPasti

• KIPasti
  • Repository organization
  • Metadata
  • Verticalized content
  • Data access
  • How to cite
  • Changelog

The KIPasti corpus is part of the larger KIParla collection, which can be freely queried through the NoSketch Engine interface.

The ParlaBO corpus was compiled within the framework of “DiverSIta – Diversity in spoken Italian” project, funded by the Italian Ministry of University and Research (MUR) (PRIN 2022 PNRR Call).

It consists of over 40 hours of spoken data collected in thirteen different Italian regions (Abruzzo, Basilicata, Calabria, Campania, Emilia-Romagna, Lazio, Lombardy, Marche, Apulia, Sardinia, Tuscany, Umbria, Veneto) during mealtime conversations, generally within family settings. The interactions, recorded between 2020 and 2024, involved 145 speakers with different origins, ages, education levels, and occupations. Italian is predominantly used in all interactions, but in most of them (78%), various passages in dialect are also present.

The transcriptions have been anonymized.

Overall, the module is made up of 63 conversations and includes 145 speakers.

Repository organization

This repository contains:

• metadata for both speakers and conversations, in the metadata subfolder (see metadata section below)
• descriptions of the set of transcription conventions used for this module (Transcription conventions)

For each conversation you will find:

• .eaf file in eaf/ folder: time-aligned Jefferson-style transcriptions (open with ELAN).
• .txt file in linear-jefferson/ folder: linearized Jefferson-style transcription.
• .txt file in linear-orthographic/ folder: linearized transcription retaining only orthographic words.
• .tsv file in tsv/ folder: verticalized version of the transcription, with Jefferson-style information decoupled from the text as features See verticalized-content for more information.

Linear files in linear-jefferson/ and linear-orthographic/ contain one Transcription Unit (TU) per line. Each line has two columns: the first is the speaker code, and the second is the transcription. TUs are sorted by their start time.

Metadata

Each participant and each conversation are associated to a series of metadata, that can be found in the metadata/participants.tsv and metadata/conversations.tsv fils. Metadata is to be interpreted as follows:

1. Participants metadata:

   • code: unique anonymized 5-char identifier for each participant. Unknown, occasional participants to conversations are associated with a special ??? code.
   • gender: either M for masculine or F for feminine
   • age-range: 5 years range including the participant’s age.
   • birth-region: Italian region¹ where the participant was born. If outside Italy, the label estero is used.
   • occupation: occupation of the participant, according to ISTAT categories. For more information see occupation label
   • study-level: highest completed level of education²
   • L1: participant's first language, can be either italian, dialect or other.
   • Additionally, the metadata/participants.tsv also contains a conversations colum that summarizes the conversations in which the participant appears.

2. Conversations metadata:

   • code: unique identifier for conversation
   • type: type of interaction, for this module all conversation are free-conversations
   • duration: duration of the conversation, expressed in hh:mm:ss format
   • participants-number: number of participants in the conversation
   • languages: languages spoken in the conversation, can be either italian or dialect, or both.
   • participants-relationship: relation, symmetric for this module
   • moderator: presence of a moderator
   • topic: free for this module
   • year: year of collection
   • collection-point: two-letter code of the collection area
   • collection-region: italian region where the collection point is located
   • macro-region: either NORTH, CENTRE or SOUTH
   • Additionally, the metadata/conversations.tsv also contains a participants field that recaps the codes of the participants to that conversation

Verticalized content

Conversations are also available in a vertical, pseudo-tokenized version in tsv/. Tokenization is obtained by validating the Jefferson transcription using custom toolsand splitting on token boundaries: whitespaces, prosodic links (=), and apostrophes used for elision in Italian orthography. Each transcription-derived token is then documented on one row.

Each token is represented as 13 columns, as follows:

1. token_id: unique token identifier within the conversation
2. speaker: speaker code as it can be found in metadata/participants.tsv
3. tu_id: progressive identifier assigned to transcription units
4. span: portion of the original jefferson transcription containing the token
5. form: orthographic form of the token. This differs from the span as special symbols are stripped out and represented as jefferson_feats. Moreover, shortpauses ((.) in the transcription) are represented as [PAUSE] and unintelligible tokens (sequences of x in the transcription) are represented as x
6. type: one of
   • linguistic: everything that is considered to be a content linguistic token
   • nonverbalbehavior used for transcribed non verbal behaviors, such as laughing or sighing
   • shortpause that identify pauses
   • unknown that identify unintelligible spans in transcription
   • error is a residual class to mark cases where the transcription is not well formed according to Jefferson format. Therefore, the token is not analyzed and transcription will be corrected in future releases.
7. variation encodes whether the transcription unit includes code mixing with dialects. In this case, all tokens in the unit have some as a value, otherwise none is used.
8. jefferson_feats: the column collects a list of word-level features derived from the transcription in Jefferson format. More specifically:
   • SpaceAfter=No: no whitespace between this token and the next (e.g., l' in l'anno)
   • ProsodicLink=Yes: a prosodic link (=) to the following token,
   • Intonation can assume values Falling, Rising or WeaklyRising and translates word final punctuation sign in Jefferson transcriptions (i.e., ., ? and , respectively)
   • Interrupted=Yes: words interrupted in speech, transcribed with final, transcribed with final ~
   • Truncated=Yes: truncated forms (e.g., anda' for andare, common in some Italian varieties)
   • Volume can assume values High or Low and translated Jefferson's uppercase and ° respectively
9. align: alignment features for the first and last token of each TU, through AlignBegin and AlignEnd features expressed in milliseconds
10. prolongations: positions of sound prolongations (colons :) within the word, encoded as a comma-separated list of <char_id>x<count> pairs
    • char_id is the zero-based index of the character in the token's orthographic form.
    • count is the number of consecutive colons immediately following that character in the original span
    • example: for span = ese::mpio:, its orthographic form is esempio and the prolongations field would assume value 2x2,6x1 (the 3rd letter e has 2 colons; the 7th letter o has 1 colon).
11. pace: marks whether the token participates in a fast or slow paced span within the word.
    • Format: Fast=<char_id_start>-<char_id_end> or Slow=<char_id_start>-<char_id_end>
    • Indices are zero-based, inclusive, and refer to character positions in form
12. guesses: character span(s) transcribed as uncertain (i.e., in round brackets in the Jefferson transcription).
    • Format: <char_id_start>-<char_id_end> (zero-based, inclusive, over form)
13. overlaps: comma-separated list of character spans participating in simultaneous speech, with an overlap group identifier.
    • Format: <char_id_start>-<char_id_end>(<overlap_id>), where indices are are zero-based, inclusive indices over form and overlap_id is the progressive number of the overlapping group within the TU
    • Examples: the span e[se]mp[i would be encoded as 1-3(2),5-6(3) meaning that characters from position one (inclusive) to three (exclusive) participate to span number 2 while the last character (with id 5) participates to the third overlapping span of the transcription unit. When the overlapping id was not decidable, a ? is used

Data access

Due to GDPR restrictions, pseudo-anonymized audio files (MP3) are available under a restricted-access license. To request access, please contact the corpus coordinators through the KIParla website and follow the provided procedure.

How to cite

To cite this module please include

Mauri, C., Ballarè, S., & Zucchini, E. (2024). Modulo KIPasti (Version 1.0.0) [Data set]. https://doi.org/https://doi.org/10.60760/unibo/kipasti.

in your references

	@misc{Mauri_Modulo_KIPasti_2024,
		author = {Mauri, Caterina and Ballarè, Silvia and Zucchini, Eleonora},
		doi = {https://doi.org/10.60760/unibo/kipasti.},
		month = oct,
		title = {{Modulo KIPasti}},
		url = {https://github.com/KIParla/KIPasti},
		year = {2024}
	}

If you use the ParlaBO module in your research, please also reference this repository (commit/tag) in your data statement or appendix.

Changelog

• 2025-10-07 v1.0.0

  • First release

• 2025-11-28 v1.1.0

  • Major fix: wrong speaker attribution in linear-jefferson and linear-orthographic
  • Minor fix: empty turns in linear-orthographic were removed

---

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

Footnotes

1. abruzzo, basilicata, calabria, campania, emilia-romagna, friuli-venezia-giulia, lazio, liguria, lombardia, marche, molise, piemonte, puglia, sardegna, sicilia, toscana, trentino-alto-adige, umbria, valle-d-aosta, veneto ↩

2. elementary-school, liceo-diploma, middle-school, phd, technical-vocational-diploma, university-degree, university-degree-ongoing ↩