identifying foreign language learning profiles in game-based environments by using data mining
TRANSCRIPT
Identifying foreign language learning profiles in game-based environments by using data mining
Manuel Palomo-DuarteAnke Berns
Andrés Yañez EscolanoJuan Manuel Dodero
Introduction
● Foreign language courses (ECTS)– few hours of language practice in class
– many hours of independent learning
● We propose using a 3D virtual world to:– Encourage independent learning
– Improve communication skills
– Provide a fun and “natural” learning environment
● But How do students benefit from the experience?– Do all benefit equally?
Design and architecture
● We used the Opensim Virtual world engine– Well-known reliable open source
– Client-server approach
Game design
● Level 1: Memory game– Pre-requisite for level 2
– Single player-mode
– Focus on changing prepositions:
● Combination with the accusative or dative
● And proper gender
– 30+ minutes
Game design
● Level 2: Hidden room– Collaborative game, two players, ~35 minutes
– Reinforce of the previously introduced prepositions
– Game aim: tidying a room using a text-chatStudent1 knows where to collocate the objects
Student2 must collocate the objects in the right place
Case study: settings
● German as a second foreign language undergraduate course, Univ. of Cadiz (Spain)– Beginner level (A1.2 CEFR), 6 ECTS
● 48 hours of face-to-face teaching● 102 hours of independent learning
– Students passed a previous course (A1.1 CEFR)
– 99 students took part in the experiment
Case study: settings
● Experiment steps:
1.Pre-test
2.Playing the game taking one role
3.Post-test 1
4.Playing the game taking the other role
5.Post-test 2
6.Technology Acceptance Model (TAM) questionnaire
Case study: analysis
● Anonymous TAM-based questionnaire [−2,+2]– Students confirm that they benefited from the game
● Interacting with others (average 1.64)● For fostering vocabulary (average 1.42)● Increasing grammatical knowledge (average 1.53)
– Students felt comfortable with the experience● Few issues using technology
Case study: analysis
● Clustering (k-means algorithm)– A well-known data mining technique
– Summarizes global behaviour in groups with similar values in the considered attributes
– Attributes of tests used for clustering:● Number of words written in a text describing the position of
different objects on a picture● Ratio of lexical errors by written word● Ratio of grammatical errors by written word
● Additionally, checked grades in previous A1.1 course
Case study: analysis
● Lexical clustering:– General constant
improvement
– C0 and C1:● Similar pre- level● Very good students● C1 students usually
wrote more words and →? improved more
Case study: analysis
● Lexical clustering:– C2: ~half of students
● Good previous oral skills● Great improvement in
post-test2
– C3: ¼ of students● Wrote few words● Best previous writing
grades● Did not learn so much
with the game?
Case study: analysis
● Lexical clustering:– General constant
improvement
– C0 and C1:● Similar pre- level● Very good students● C1 students usually
wrote more words and →? improved more
Case study: analysis
● Grammatical clustering:– General improvement only after post-test2
– C2: ~half of students, good at participation● Outstanding final results
– C0, C1 and C3: same final level● C0 and C3 had same poor pre-test level
– C3 students wrote always an average number of words and they started improving in post-test1
– C0 students only wrote a reasonable number of words in post-test1 and improved only after post-test2. Needed a second round?
● C1: students between C0 and C1, good previous participation grades →? not comfortable with the game
Case study: analysis
● Lexical and grammatical clustering (5 groups):– C3, 40% of students: very good pre-test results and
the best final results (more words than average)
– The other clusters have similar number of students:● C0 and C4: good at previous tasks
– C0 good at oral tasks and average writing skills in post-test2– C4 made effort to write more, but worst post-test2
● C1 and C2: similar previous performance– C1 poor in vocabulary, good at grammar– C2 the opposite
Conclusions
● Students felt conformable with the game– Although some decreased grades from previous semester
● Particular insights:– Previous oral skills ~ influence in vocabulary
– Good (but not brilliant) student in previous course + good participation got best results in grammar
– Scarce relation to previous grades in mixed analysis
– Some students needed two rounds to improve
● Future work:– Analysis including game-chat logs
– Applying other clustering techniques or association rules