Download: LTfLL-Service-Description

Problem: Educational institutions are starting to widen their offerings to a large number of lifelong learners. Traditionally, assessments to position learners use essay writing, questionnaires, multiple choice tests, or simple oral examination. Due to an increased diversity in learners’ educational backgrounds, education providers have started to use online learner interactions as part of a wider portfolio analysis to assess the learner position and to enable personalised learning. The growth in the number of registrations and more complex positioning threaten to increase the workload of tutors to unmanageable levels.

Solution: The LTfLL positioning services perform a qualitative and a quantitative analysis of learner texts (= ‘knowledge poor approach’). Qualitative analysis involves the scoring of phrases extracted from learner texts according to distinctive features of their usage by comparing its frequency in high and low quality texts as graded by experts. The output of this analysis is based on the learner’s written phrases and not simply on word frequency. Users can inspect the scored phrases visually. Quantitative analysis uses information such as occurrence counts of these phrases to compute a measure of fit of the learner language as compared to the relevant CoP. Users of the positioning service should interpret the qualitative and quantitative results to assess the learner’s position and to decide what course units the learner needs to study, and where additional support might be needed. The services need fine tuning for each CoP by training them on representative texts (e.g. textbooks, highly graded peer texts, etc.). This builds up into a 'reference corpus' which is used to infer the set of relevant concepts. To help experts in building that corpus, the service statistically analyses instructional text materials and suggests which of those should be added to the corpus. Corpus texts then are classified by area of expertise and grading. With these texts the quantitative output for each learner text is generated by measuring the distance between a text and all texts of the 'reference corpus' in two vector space models, i.e. bag of words and bag of phrases. In addition to the ‘knowledge poor approach’, conceptual coverage of learner texts is computed using an ontology by counting how many associated concepts are found in the learner texts (= ‘knowledge rich approach’). As output the percentage of covered relevant concepts will be presented. The analysis of conceptual coverage of learner texts involves the use of an ontology and lexicalisations of concepts belonging to that ontology (e.g. phrases extracted by means of the qualitative analysis) to count how many relevant concepts are found in the learner texts. Finally, after examining the service output, users can examine the appropriate list of instructional texts from the reference corpus. Tutors can use results to decide which materials need to be studied by the learner, and in which area of conceptual knowledge the learner may require further support. Learners can evaluate their own position and identify their strengths and weaknesses.