Natural language generation of trend descriptions for pedagogic purposes

2023 Fiscal Year Research-status Report

• Project/Area Number: 22K00792
• Research Institution: The University of Aizu
• Principal Investigator: BLAKE John 会津大学, コンピュータ理工学部, 上級准教授 (80635954)
• Co-Investigator(Kenkyū-buntansha): Pyshkin Evgeny 会津大学, コンピュータ理工学部, 上級准教授 (50794088)
• Project Period (FY): 2022-04-01 – 2025-03-31
• Keywords: NLG / trend descriptions / describing graphs / intelligent CALL / data series description / NLP / language generation

Outline of Annual Research Achievements

We made use of the corpus of trend descriptions to extract prototypical rhetorical patterns of trend series descriptions and the relative frequency of functional exponents that are used to realize these.

We extended the codebase of the description generator, which is now able to generate trend descriptions at five proficiency levels from beginner through to upper intermediate using rule-based parsing of a spreadsheet. As the proficiency level increases so does sentence complexity, grammatical variety and vocabulary range. Users can switch between levels to see how the trend description changes with language proficiency. Our next step is to explore the use of a large language model to generate suitable descriptions for advanced language learners.


We have also developed a work flow to streamline AI-generated video explanations to accompany the generated texts.

Current Status of Research Progress

1: Research has progressed more than it was originally planned.

Reason

Progress on the software has exceeded our initial expectations, and even have also developed a smoothing algorithm that reduces the number of data points to enables a description to be created even if there are hundreds or thousands of data points.

Strategy for Future Research Activity

We expect to place the prototype online in the next few months, and conduct experiments on its accuracy, usability and efficacy.