New version of our open-source Wikipedia2Vec tool has been released
What We Do
We are an AI startup focused on developing natural language processing to enhance intelligent AI for all organizations.
Large Language Models
LLMs are used ubiquitously in natural language processing. LUKE, a large language model that is one of our research results, has shown state-of-the-art performance in multiple natural language processing benchmarks and has been included in the Transformers library. The widget at the top of this page visualizes the token embeddings of multilingual LUKE.
Question answering is a basic way for humans to interact with AI using natural language. At NIPS 2017, our AI outperformed all other AIs and won by a large margin against the human U.S. quiz champion team. In the standard question-answering benchmark SQuAD v1.1, we achieved state-of-the-art performance in 2020. In the NerurIPS 2020 question answering competition, we achieved performance competitive with Microsoft and Meta AI.
ANSWER ROBOT allows users to create AI bots instantly from FAQs. It can answer with high accuracy without any human effort.
QA ENGINE is an award winning question answering system that automates customer helpdesks.
SMART ANNOTATOR is a text annotation tool that is based on our AI to cluster text by its meaning.
Aiming to enhance data value in business intelligence
Add-on high accuracy QA on Japan’s No.1 Chat support tool
Introducing our pre-trained question answering service Answerrobot
Demonstrating the possibilities of utilizing the vector representations of entities to improve multilingual pretrained language models
Utilizing the vector representations of entities to improve multilingual pretrained language models
Launch an AI bot in Microsoft Teams
Powered by the world’s first pre-trained Japanese QA model Sōseki
Vespa is the former AI division of Yahoo!
Our two papers were accepted at a top NLP conference