A mixed-method approach to normalising Dr. Math microtext

摘要

Dr. Math is a South African math tutoring service, which allows high school learners to use a mobile phone-based chat application to contact volunteer tutors. The learners structure their queries in a form of microtext, which the tutors find difficult to interpret. This study presents an automated process as a means to normalise a Dr. Math microtext-based learner query, so that it more closely resembles English. The process consists of a variety of replacements, based on known short forms, pre-generated English-variants and phonetic similarity. Input queries are processed at a word-level, regardless of underlying grammar. The normalisation process is validated in two steps. The first step uses spell-checking to determine that the process improves the overall spelling accuracy, of the original microtext-based input queries, in both test and validation conditions. Krippendorff's α is used in the second validation step to calculate the level of agreement, with regards to the legibility of normalised text, between two human coders. The α value of 0.681 is deemed sufficient for exploratory research. The coders agree that the normalisation process yields an improvement in legibility.