JILDA
JILDA (Job Interview Labelled Dialogues Assembly) is a novel collection of chat-based mixed-initiative, human-human dialogues related to the job offer domain and for the Italian language.
JILDA dialogues were collected using a Map-Task approach which involved 50 Italian native speakers to simulate. These speakers simulated a chat-based conversation between a ”navigator” and an applicant. In this way we collected in this way 525 mixed-initatives dialogues, with 15-20 Navigator-User interaction.
This repository includes also the MTurk dataset, a collection we previously built for the same domain and language. The MTurk dataset includes 220 template-based dialogues collected via Amazon Mechianical Turk.
The two datasets were annotated using the same annotation schema. Below are defined the main dialogue acts and slot used to annotate the dialogues collected:
Dialogue acts
All the DAs can be referred both to the applicant’s utterances (aka usr) and to the navigator’s utterances (sys):
- greet,
- inform_basic,
- inform_proactive,
- request,
- select,
- deny
Slots
They represent the attributes of the entity job-offer:
- job_description,
- contract,
- duties,
- skill,
- past_experience,
- degree,
- age,
- language,
- area,
- company_name,
- company_size,
- location,
- contact,
- other
Cite
Irene Sucameli, Alessandro Lenci, Bernardo Magnini, Maria Simi, and Manuela Speranza. 2021.
Toward Data-Driven Collaborative DialogueSystems: The JILDA Dataset. In IJCOL
People
- Irene Sucameli - University of Pisa
- Alessandro Lenci - University of Pisa
- Bernando Magnini - FBK
- Maria Simi - University of Pisa
- Manuela Speranza - FBK