Dyslexia is a learning disorder that affects around 10-20% of the world's population. In Denmark, an official nationwide diagnosis of dyslexia, Ordblindetesten, is used, which is designed to reveal phonological decoding difficulties. Technological diagnostic methods are under development internationally, and among these, eye tracking is particularly interesting, as it provides access to rich information about cognitive processing. The purpose of this study is to assess the extent to which machine learning classification of eye tracking data can be used as a diagnostic tool for dyslexia in Denmark. The data collection consists of eye tracking recordings from dyslexic and non-dyslexic readers during natural reading of Danish texts. Here we use an in-depth psycholinguistic analysis where the reading patterns from the two groups of participants are compared. The results show a significant difference in selected measurable eye tracking features. Based on this difference in the data between the two groups of readers, we selected machine learning classifiers among which the best performing model was a Random Forest classifier. This scored an accuracy of 85%, indicating that although there is still some uncertainty, our methods can detect measurable differences in eye movements in dyslexic and non-dyslexic readers, and can thus potentially be incorporated into a screening process.

Ordblindhed er en indlaeringsforstyrrelse, som påvirker omkring 10-20 % af verdens befolkning. I Danmark benyttes en officiel landsdaekkende test til at diagnosticere ordblindhed, Ordblindetesten, som er lavet til at afdaekke fonologiske afkodningsvanskeligheder. Samtidig udvikles flere teknologiske diagnosticeringsmetoder på internationalt plan, og blandt disse er eye tracking specielt interessant, da denne metode giver adgang til mange informationer om kognitiv processering. Formålet med denne undersøgelse er at vurdere, i hvilket omfang maskinlaeringsklassificering af eye tracking-data kan bruges til at diagnosticere ordblindhed i Danmark. Dataindsamlingen består af eye tracking-optagelser af ordblinde og ikke-ordblinde laesere under naturlig laesning af danske tekster. Her benytter vi en dybdegående psykolingvistisk analyse, hvor laesemønstrene fra de to grupper af deltagere sammenlignes. Denne viser en signifikant forskel i udvalgte målbare eye tracking-komponenter. På baggrund af denne forskel i datasaettet mellem de to grupper af laesere udvalgte vi maskinlaeringsmetoder til at klassificere laesemønstrene. Dette resulterede i en nøjagtighed på 85 %, hvilket indikerer, at selvom der stadigvaek er en vis usikkerhed, er vores metoder i stand til at opdage målbare forskelle i øjenbevaegelserne hos ordblinde og ikke-ordblinde laesere, og de kan dermed på sigt taenkes ind i udredningsprocessen. EMNEORD: eye tracking; maskinlaering; ordblindhed; laesevanskeligheder; psykolingvistik INTRODUKTIONOrdblindhed (eller dysleksi) er en specifik indlaeringsforstyrrelse, som angiveligt påvirker omkring 10-20 % af verdens befolkning (Rello & Ballesteros 2015, Kaisar 2020). Kendetegn for ordblindhed inkluderer besvaer ved laesning, skrivning samt ordafkodning og er ikke relateret til intelligens (Perera et al. 2018, Rauschenberger et al. 2017. Det er vigtigt at opdage ordblindhed så tidligt som muligt, da vanskelighe-

Eye movement recordings from reading are one of the richest signals of human language processing. Corpora of eye movements during reading of contextualized running text is a way of making such records available for natural language processing purposes. Such corpora already exist in some languages. We present CopCo, the Copenhagen Corpus of eye tracking recordings from natural reading of Danish texts. It is the first eye tracking corpus of its kind for the Danish language. CopCo includes 1,832 sentences with 34,897 tokens of Danish text extracted from a collection of speech manuscripts. This first release of the corpus contains eye tracking data from 22 participants. It will be extended continuously with more participants and texts from other genres. We assess the data quality of the recorded eye movements and find that the extracted features are in line with related research. The dataset available here: https://osf.io/ud8s5/.