One of the challenges for computer aided language learn-ing (CALL) is providing high quality feedback to learners. Anobstacle to improving feedback is the lack of labelled trainingdata for tasks such as spoken ”grammatical” error detection andcorrection, both of which provide important features that canbe used in downstream feedback systems One approach to ad-dressing this lack of data is to convert the output of an auto-matic speech recognition (ASR) system into a form that is closerto text data, for which there is significantly more labelled dataavailable. Disfluency detection, locating regions of the speechwhere for example false starts and repetitions occur, and subse-quent removal of the associated words, helps to make speechtranscriptions more text-like. Additionally, ASR systems donot usually generate sentence-like units, the output is simplya sequence of words associated with the particular speech seg-mentation used for coding. This motivates the need for auto-mated systems for sentence segmentation. By combining theseapproaches, advanced text processing techniques should per-form significantly better on the output from spoken languageprocessing systems. Unfortunately there is not enough labelleddata available to train these systems on spoken learner English.In this work disfluency detection and ”sentence” segmentationsystems trained on data from native speakers are applied to spo-ken grammatical error detection and correction tasks for learn-ers of English. Performance gains using these approaches areshown on a free speaking test.