AN ALGORITHM ANALYZING PHONEME-GRAPHEME AWARENESS THROUGH THE BREAKDOWN OF WORDS Zachariah Clifford Micallef 1 , Mark Bugeja 2 , Dunstan Briffa 1 and Dylan Seychell 2 1 Saint Martin's Institute of Higher Education, Schembri Street, Hamrun, HMR 1541, Malta 2 University of Malta, Msida MSD 2080, Malta ABSTRACT Phoneme awareness and orthography are core language skills, in this paper the relationship between both is monitored. There are two main parts to this experiment. The first is a group of algorithms that enable the measures of the described relationship and the second uses the previous measurements to solve a classification problem that related to a real-world problem. Participants are asked to spell a list of words by audio and divide each word into groups of letters that describe the shortest units of sound. This is done so that the correlation between how a person hears a word and how they write it can be observed. Once the data is collected the former part is used to deduct a score. The latter part of the experiment is done to prove the effectiveness of the scoring. Dyslexia is a clinical issue that is known to affect phonological awareness that relates to poor orthography. For this experiment, participants with this profile were asked to participate. Using the score obtained from the data, a classification model is trained in an attempt to classify dyslexic and non-dyslexic participants. This will be used to gauge the effectiveness of the scoring. The experiment has already been proven to work with a limited amount of data. KEYWORDS Accessibility, Dyslexic, Analyzing Spelling, Machine Learning, Phonological Awareness, Diagnosing 1. INTRODUCTION When observing a participant's attempt to match each phone (the shortest unit of sound) to a group of letters lead to the understanding one's phoneme to grapheme interpretation of a word. This indicates how text is being interpreted. In this paper, the awareness of how phonemes and graphemes to each other is referred to as phoneme-grapheme awareness. Having a measurement of such a skill can indicate how a language as a whole is being decoded rather than recalled. Creating a scoring method that relates to phoneme-grapheme awareness can indicate that some comprehend sounds relate to spelling more than others. Phoneme-grapheme awareness differs between dyslexics and non-dyslexics. Therefore, a classification model with the application of the scoring method should distinguish dyslexics and non-dyslexics, confirming that the scoring matches a real scenario. Previously this experiment was carried out in a primitive manner since the data was interpreted manually. This consisted of two groups of 8 participants, one consisted of dyslexic participants and the other was a control group. The reason for this was to determine if such a process is useful and effective before developing an automatic algorithm. Positive results were obtained with the use of a KNN, further research was deemed to be useful. Figure 1. Phonological Accuracy Vs Total spelling errors shows the results of the above-mentioned experiment. This paper covers how the computerized algorithm works. Testing is being undertaken on a larger scale with the use of a mobile and desktop application, which was specifically designed for this experiment. 18th International Conference e-Society 2020 155