[4Pin1] インタラクティブ(2)

[4Pin1-50] Generating Spatial Semantic Descriptions of Parking Instructions for Self-driving Cars

〇Akari Inago¹, Hiroshi Tsukahara², Ichiro Kobayashi¹ (1. Ochanomizu University, 2. Denso IT Laboratory)

Keywords:Natural Language Processing, Parsing, Automatic Driving

This paper describes a method to transform a combinatory categorial grammar(CCG) parse tree into a spatial description clause(SDC) to be grounded onto the objects or the phenomena in the real environment. We especially designed the CCG for handling spatial meanings contained in parking instructions for self-driving cars. In previous studies, we developed a parser for generating parse trees of parking instructions according to the CCG we defined. We propose a rule-based converter to perform a conversion from these parse trees into SDCs using out heuristic knowledge gained from our analysis on a careful investigation on individual examples. We built the converter and evaluated it on the results of our parser. The result of conversion shows fairly high accuracy for parse trees generated by our parser from parking instructions in the corpus used in this study.