By querying the KIParla corpus, the user declares and accepts that:

  • the consultation is carried out exclusively for scientific research purposes and without commercial gain;
  • in case of use and/or reproduction of the linguistic data contained in the resource, the authorship of the resource will be acknowledged by citing it appropriately (see below);
  • the data obtained from the consultation will not be modified (unless explicitly declared by the user);
  • the data included in the KIParla corpus will not be included in any other resource (except by agreement with the administrators).

The data from the KIParla corpus downloaded via the NoSketch Engine platform are shareable under the Creative Commons BY-NC-SA 4.0 license.

ATTENTION! 
Researchers wishing to access the audio files for research purposes must contact the resource managers via their institutional email addresses.
It is not possible to download the corpus audio files and/or record them using personal means, due to reasons related to the release form used during data collection and the European regulation on personal data protection (GDPR).
Each data collection involves establishing a relationship of trust between researchers and speakers, without which the resource itself would not exist and, above all, could not be made accessible to the scientific community.
For ethical and legal reasons, therefore, anyone consulting the corpus is required to respect the access procedures indicated by the resource managers.

Searches within the corpus are performed using the NoSketch Engine interface. It is possible to query the three corpus modules separately (KIP, ParlaTO, KIPasti, or ParlaBO) or jointly (KIParla), by selecting the preferred option from the dropdown menu at the top.

To access the corpus, click on the red button:

To access the audio files, please fill out the form and register.

You will receive an email to confirm your registration (check your SPAM folder!) and then a second email with your username and password to listen to the KIParla voices from your browser.

By registering, you will automatically be subscribed to the KIParla corpus newsletter.

 

How to cite the corpus

To cite the KIParla corpus, please use the following reference:

Mauri, Caterina, Silvia Ballarè, Eugenio Goria, Massimo Cerruti & Francesco Suriano, (2019) “KIParla corpus: a new resource for spoken Italian”. In: Bernardi, Raffaella, Roberto Navigli & Giovanni Semeraro (eds.), Proceedings of the 6th Italian Conference on Computational Linguistics CLiC-it.

KIParla Corpus Repositories

Updated documentation and the latest versions of the corpus are available on GitHub.

The description and transcripts of the KIParla corpus modules have been deposited on CLARIN-it, where the latest stable and revised version of each module can be found. Specifically, the following are accessible:

  • metadata related to both speakers and conversations,
  • the description of the set of transcription conventions used for each module
  • for each conversation: .eaf files in the eaf/ folder (transcriptions aligned with audio, using Jefferson conventions); .txt files in the linear-jefferson/ folder (linearized transcription using Jefferson conventions); .txt files in the linear-orthographic/ folder (linearized orthographic transcription); .tsv files in the tsv/ folder (tokenized version of the transcription).

Useful information for conducting a search

KWIC view: what is shown in the left and right context, around the keyword?

The surrounding text displays the intonational unit, and more generally the context, in which the searched element appears; or, more precisely, where the occurrence of the searched element appears. The context may include utterances produced by the same speaker or by other speakers involved in the same recording. Intonational units are separated by the // symbol. To identify the author of each displayed intonational unit, it is necessary to consult the transcription of the entire interaction. To view the complete transcription, select the conversation code (e.g., TOD2015) that appears to the left of the occurrence, where there is a link to the full conversation transcription.

What is displayed if I click on the occurrence?

By clicking on the occurrence, you can view the broader context in which it appears.

What is displayed by clicking on the first occurrence of ‘casa’
What information do I get if I click on the code to the left of the occurrence?
  • conversation code
  • link to orthographic transcription of the entire conversation, where I can find the occurrence using the Ctrl+F function. Each speaker is identified by the same color.
  • link to simplified conversational transcription of the entire conversation, where I can find the occurrence using the Ctrl+F function. Each speaker is identified by the same color.
  • link to audio of the conversation segment identified by the occurrence.
  • speaker-related metadata (gender [f=female, m=male], age group, region where they attended high school, occupation [s=student, p=professor], other conversations the speaker participates in).
  • conversation-related metadata (conversation type [lectures, exams, free conversations, student office hours, semi-structured interviews], collection point [bo, to], speaker relationship [symmetric vs. asymmetric], number of participants, year of recording)
What is displayed by clicking on the conversation code
I tried listening to the audio segment of a conversation but I cannot find the occurrence I was looking for. Why? What can I do?

Once the player is open, the track starts 3 seconds before the intonational unit where the searched form is located. It is still possible to navigate the player using the cursor or shift + directional arrows (for more precise control); additionally, you can refer to the full conversation transcription (see above).

What search filters can I use?
  • Query types: ‘simple’, ‘phrase’, ‘word’, ‘character’, CQL’
  • Context: it is possible to filter the results based on the co-occurrence of a word with other words in the right and/or left contexts.
  • Text types: it is possible to search for an occurrence using various metadata as filters, related to speakers (e.g., age group, profession, etc.) and/or to the conversation (e.g., relationship between participants, year of collection, etc.).
How can I save the search results?

Once you have done your search, click on the download button. At this point, select the format in which you want to save the data.

Are there any 'wild cards' I can use for searching?

Yes, a brief list is provided below:

? = Represents 'any character' and is used at the beginning and end of a word.

For example, if you want to search for all inflected forms of the past participle of the verb 'to be', you should type 'stat?' in the search field (caution! Results will obviously include, for instance, stato/i when used as nouns!).

* = Represents an indefinite number of characters.

For example, by searching for 'ca*a', I will find words that start with 'ca' and end with 'a', formed by a variable number of letters, such as carla and candidatura.

^ (in the 'word form' query type) = Represents 'any character except the one specified'.

For a more detailed guide to the regular expressions you can use, please read the Sketch Engine guide by clicking here. 

How were the interjections transcribed?

Below is a brief list of how some interjections and discourse markers were transcribed in an attempt to standardize their occurrences:

  • vabbè (and not vabbe, va beh, va be, va be’, …);
  • cioè (and not cè);
  • mah (and not mha);
  • ehm (and not emh, hem);
  • eh (and not he);
  • mh (and not m, hm);
  • hi (and not ih).