A method is proposed in which nondefined Chinese characters may be uniquely classified thus making them compatible for machine translation. An optical-digital device is used to locate nondefined geometric shapes within Chinese characters via spatial filtering techniques and cyclic cross-correlation. Seventeen nondefined geometric shapes are found in a 98 character sample which is representative of a ten thousand Chinese character vocabulary. Preliminary results indicate that 60 percent of the geometric shapes are identified exactly and over 90 percent can be identified within specified tolerances. Geometric shape energy content is shown to be pertinent in the decision process. (Author)