Overview
Our goal is to develop products that appear like toys but are highly functional. We believe that a sense of playfulness can drive significant progress in society. At Functional Toy Manufacturing, we aim to contribute to society through this approach.
In the process, we leverage mathematical principles to create toys. Specifically, we utilize contemporary machine learning techniques, such as transformers, and functional programming, which we believe will form the foundation of advanced AI. Our aim is to introduce intellectually stimulating toys that are enhanced with such specialized knowledge.
Among our creations, we are particularly proud of our latest intellectual toy, the Text Transformer(PPL++).
Introduction：Text Transformer(PPL++)
The Text Transformer (PPL++) is a tool designed to transform text, making it easier for readers to comprehend its content . The more complex the text, the more effective PPL++ becomes, demonstrating its unique capability. While it may sound ambitious, PPL++ is grounded in the essence of cognition and language, making it particularly adept at handling complex texts that adhere closely to these principles. These core principles of cognition and language are simple yet profoundly rich upon closer examination. (Principles Explained Here(Under Construction))
Texts transformed by PPL++ will likely be unlike anything you've encountered before. Though it may initially seem like a toy, it is remarkably practical. (Product Details Here)
Exploring the Potential of Language
We are dedicated to exploring how far we can unlock the potential of natural language. Currently, written text is often aligned with spoken language (which is linear), but it isn't optimized for the act of reading (which is multi-dimensional). However, with digital documents, we have the flexibility to alter the display as needed.
Our current best solution is PPL++, but we are open to further evolving it or taking a completely different approach. It is also intriguing to consider revising rigid grammatical rules to better optimize text display. This could lead to more precise expressions and the handling of abstract concepts more effectively.
As of 2024, interactive AIs like ChatGPT are at their peak. It is remarkable how accurately they can respond to highly abstract questions. One reason for this is that the language used in their training data already follows certain logical principles. We believe that by evolving language to allow for more precise expressions and abstract concepts, AI will also advance.
There are also areas where complex texts are rarely written due to their difficulty in comprehension, leading to a lack of training data. By using PPL++, we can simplify the interpretation of such complex texts and increase the available training data.
We believe that by expanding language, we can also expand the means of acquiring knowledge. Thank you for your attention.
Affiliations
Nagoya Chamber of Commerce and Industry、 Nagoya Innovator’s Garage（Nagoya City, Central Japan Economic Federation）、The Association for Natural Language Processing, Member
Member
Taichi Ishikawa
Biography
Taichi Ishikawa graduated from Kyoto University Faculty of Law, and then from Nagoya University’s Faculty of EngineeringNagoya University’s Faculty of Engineering, Department of Physical Engineering, and Graduate School of Engineering (Nuclear Engineering Major). He worked at Shinko Electric Industries Co., Ltd. in Nagano, where he was involved in development, legal affairs, intellectual property (patents), and equipment development. In January 2022, he founded Functional Toy Manufacturing and continues his work there today.
Interests
- Hot springs, walking, meditation
- Nagano, Nagoya, Kyoto
- Yoga, Latin
- JavaScript, Python, Haskell
- Type theory and logic, machine learning (Transformer, GNN)
Development Background
Throughout my life, I have repeatedly challenged myself with difficult texts, only to encounter frustration and setbacks. During my university years, I attempted to read philosophical texts but was overwhelmed by the complex terminology. I even considered law school, only to be deterred by the similarly challenging legal language.
Eventually, I pursued a career in engineering, where I found some success in my field. However, when I sought to expand my knowledge in mathematics and theoretical physics for personal enrichment, I struggled with the equations, symbols, and jargon that made reading the literature daunting.
When I later joined a semiconductor components company in Nagano, working in the legal and intellectual property departments, I again faced complex texts—this time in the form of contracts, laws, and patent documents. Unlike philosophy or law, though, the patent industry already had tools to tackle these challenges: multi-color highlights and keyword highlights.
In the context of patent searches, documents are often highlighted in different colors for each search term. For example, in the diagram, terms like "graph" and "neural network" are highlighted, while frequently appearing words such as "section" and "number" are pre-registered and highlighted. This keyword highlighting helps easily identify where search terms appear in the text.
However, the same color-coded approach also allows for a different reading experience. I wondered if this multi-color highlighting could be applied to the complex texts I had previously struggled with .
So, I began developing this idea as a side project, separate from my work duties, to avoid any conflict of interest. After three years of part-time development, I decided to focus on it full-time and left my job to pursue it independently.
Initially, I had no working prototype, no customers, and lacked some of the technical skills needed to complete the prototype. With the help of many collaborators, I managed to overcome these challenges over three years. Now, the prototype is complete, and the difficult texts that once challenged me are much easier to read.
For instance, I have applied PPL++ to machine learning textbooks and papers, technical articles on the web, device manuals, and library documentation, all of which have become significantly more readable. However, I have not yet tested it in the use case I originally intended—my student self. Nowadays, I am more in the phase of utilizing knowledge rather than voraciously acquiring it, and the intense hunger for knowledge has diminished somewhat.
To be honest, I cannot claim that this tool has dramatically transformed my reading experience as a student would have found it indispensable. My enthusiasm for reading every book in the library has waned, and I can only speculate that my younger self would have eagerly embraced this tool.
Nevertheless, I remain convinced that this technology is fundamentally sound, rooted in the essence of language, as explained on this website.
Going forward, I want to test this technology's effectiveness with others. If it proves beneficial, I aim to popularize it and contribute to society. This technology should help with difficult texts that traditional reading methods struggle to address, significantly contributing to scientific and technological progress.
This technology is also well-suited for verbose yet precise texts like those produced by ChatGPT. As AI becomes more integrated into education, I find myself spending more time learning from AI, using this technology to facilitate smooth interactions.
I hope this tool will be useful in various scenarios, especially for those in the knowledge-acquisition phase of their lives. This is where the development of this tool began.