Introduction

The DigiSami project (www.helsinki.fi/digisami/) aims to study the effect of digitalisation on small Finno-Ugric language communities and to support visibility and revitalisations of the endangered languages by creating digital content as well as developing language and speech technology tools, resources, and applications that can be used for automatic speech and language processing. The project focuses especially on the North Sami language, and explores various spoken language issues (speaker identification, multimodal conversation analysis, laughing), with the challenging goal of demonstrating viability of an interactive dialogue system in the North Sami language, SamiTalk, following the multilingual open-domain robot application WikiTalk. WikiTalk is an interactive robot application that enables users to find out more about subjects that interest them by discussing with the humanoid robot. They can navigate through the Wikipedia articles, ask for more information on interesting subjects, and get the robot to read the related Wikipedia article for them.

Data collection

The project organsied Sami language data collection and Wikipedia article writing through series of community events. The participants were invited to take part in three different tasks: discussion and writing Wikipedia articles, reading aloud of existing Wikipedia texts, and taking part in a free conversation which was video recorded. The events took place in the central Sami speaking areas, selected so that they represented different North Sami dialects. The locations were three villages in Finland: Utsjoki (Ohcejohka), Inari (Anár) and Ivalo (Avvil), and two villages in Norway: Kautokeino (Guovdageaid) and Karasjoki (Kárásjohka). (see more of the DigiSami data and data collection in). Text readings and conversations were recorded by EDIROL R4Pro four-channel recording device with AKG 417 L-microphones. Two Panasonic HC-X920 video cameras and three GoPro HERO3 cameras were used for video-recordings. Conversational speech was also recorded by the cameras own microphone. The reading took place in a calm and normal speaking manner, and the participant could study the articles in advance. The conversations were between two or three people, and the participants were instructed to discuss freely about their own interests or about the Wikipedia articles they were to write (e.g. Sami language, Sami costume, music, reindeer herding, and snow- mobiles). The topics vary from everyday life (next vacation, driving school, cars) to translation between Sami and other languages and to technological tools that have been made to help writing North Sami more correctly. The conversations differ in style: familiar participants have casual conversations and they often refer to things they had been talking about earlier. Conversations between a pupil and a teacher are more formal and resemble interviews rather than conversations.

Dataset statistics and organization

There were 28 participants, 10 men and 18 women. Their age range from 16 to 65 years: 17 were 16-21 years old, five 30-44 years old, and six 49-65 years old. The participants were native speakers of North Sami, and almost all (26) reported using North Sami daily; one participant reported using North Sami weekly and one participant monthly. All participants were bilingual, and spoke either Finnish (Utsjoki, Ivalo, Inari), or Norwegian (Kautokeino and Karasjoki). Most participants had lived their life in the Spmi area, although not in the same place. Ten participants had also lived in bigger cities in the southern part of the area, Oulu and Rovaniemi in Finland and Bergen in Norway, for a short period of time.

Contents and annotations

The dataset provides recording in two primitive types of data: audio and video, which are annotated for three different tasks:

• The transcriptions are provided in both North Sami, and translated English.
• The laughter annotations mark the laughing events from each speaker.
• The topic annotations, in English, specify discussing topic along the conversation.

For more information concerned the structure of the dataset, you can check the Readme.txt included in the dataset.