Page 37 - Kaleidoscope Academic Conference Proceedings 2021

from the environment and how. This finding is important in
terms of supporting the theory of natural pedagogy.

Parent-infant communication promotes brain development
of the infant as explained above. It further affects subsequent
vocabulary growth of children. A child-vocabulary-
development database was developed for elucidating
language acquisition in infants. Built upon this database, a
system called “Pitarie” (pronounced pitari-eh) was
developed (Figure 4) that enables the user to search for
picture books that match the interests and developmental
stage of individual children [5]. Pitarie has been Figure 5 – Personalized picture books
commercialized by NTT DATA Kyushu, and it was installed
at the Fukui Prefectural Library in April 2019.

4. ACHIEVING NATURAL CONVERSATIONS
BETWEEN AI AND HUMANS

Conversational AI involves a computer agent chatting with
people (in chat-oriented dialogue systems) and providing
information and guidance (in task-oriented dialogue
systems). In both cases, to achieve natural and meaningful
conversations between AI and humans, it is necessary to
advance and integrate both natural language processing and
knowledge processing technologies. The English written
exams administered by the National Center for University
Entrance Examinations are good benchmarks to test the level
of such technologies for conversational AI. NTT has been
participating in the Todai Robot Project called “Can a robot
get into the University of Tokyo?” launched in 2011 by the
Figure 4 – Pitarie: a system that searches for National Institute of Informatics, thus tackling English
picture books exams in collaboration with its research partners. The
developed AI system based on deep learning achieved a
More recently, “personalized educational picture books” surprisingly high score of 185 out of 200 points (64.1 T-score)
(Figure 5) were proposed in collaboration with NTT Printing in the exam’s 2019 version [6].
Corporation. These picture books are personalized to each
child so that the next new words for the individual child to The conversational AI research at NTT includes developing
learn are selected and inserted into the books. Those new a unique conversational AI for role playing called Narikiri
words are predicted from the child’s current vocabulary, AI, which reproduces the behaviors of a celebrity or a
which is given by the child’s parents as a simple vocabulary character in a novel or a game [7]. Its dialogue data is
check list, and the child-vocabulary-development database. collected through sets of questions and answers posted by
This venture began with picture books to be read aloud to online users, where one user asks a certain character a
children (left and middle books in Figure 5), but more recent question, and another user mimics the character’s personality
research revealed that an understanding of characters and to answer it. NTT DOCOMO developed and commercialized
their correspondence to sounds actually starts around three “character chat technology” based on this AI and released
years old, slightly before the ability to read and write “AI Jimmy.” A new conversational AI project has started in
hiragana and katakana (Japanese syllabic characters). Thus cooperation with Seika Town, Kyoto, involving its official
“names-in-hiragana/katakana picture books” were proposed public-relations character (mascot) Kyomachi Seika as
to generate interest in characters targeting children of about shown Figure 6. The mascot character has attracted many
three years old (right book in Figure 5). Personalized fans through social networking services and is a perfect
educational picture books can now be ordered online at match with Narikiri AI. In this project, dialogue data,
https://ehon.nttprint.com/. Books as tangible printed including a large amount of knowledge and experience about
materials, not digital ones, that promote full interaction Seika Town, was collected through cooperation with the
evoking all five senses are significant for children.
residents of the town to build a dialogue system to provide
appropriate information and guidance regarding tourism and
town administration as well as to flexibly answer user
questions and requests [8].

– xxxiii –