Research in Intelligent Computer-Assisted Language Learning (ICALL)
integrates Natural Language Processing into Computer-Assisted Language
Learning. For such research to be innovative and sustainable, we
believe that it needs to combine Second Language Acquisition (SLA) research,
precise linguistic modeling, and sound computational linguistic
methods (Meurers 2012; Meurers & Dickinson 2017). On this
foundational research, we ground the development of digital tools
addressing real-life education needs, in school and lifelong learning.
-
Evidence for Learning
We develop methods analyzing learner language to broaden the empirical
evidence for development, both in terms of linguistic
constructions and general linguistic complexity, including task
effects and L1 transfer.
Analyzing Learner Language
-
Intelligent Language Tutoring
We create interactive systems that support foreign language learners
in practicing language skills with incremental, scaffolding feedback
— like a human tutor would, who unfortunately we can't always have
around.
Interactivity
-
Enriching Input
We design search engines and linguistic complexity measures
needed to identify the input that best fosters learners in their
language development. We also check whether educational materials
are adapted to their audience.
Adaptivity
About us
Detmar Meurers
started the ICALL research group at the Ohio State University, where
he was a faculty member at the Department of Linguistics from 2000 to
2008, before moving to the University of Tübingen.
Complementing
the current
members of our group, some former members of the group
include Xiaofei
Lu, Markus
Dickinson, Luiz
Amaral, Martí
Quixal, Robert
Reynolds and Sowmya
Vajjala.
We pursue our interdisciplinary agenda as part of
the LEAD Graduate School
and Research Network in Empirical Educational Science, in close
collaboration with the LEAD Distinguished International Professor
Patrick Rebuschat. We
are founding members of
the Heritage
Language Network, the
INDUS DFG
Network, and collaborate with our linguistics colleagues in
the SFB
833.
Analyzing Learner Language
In
the SFB 833-A4 project,
we are developing automatic meaning assessment methods for short-answer
reading comprehension.
To collect a rich task-based corpus in a real-life teaching context,
we created the WELCOME app (Ott et al., 2012) and obtained the CREG
corpus (36k answers to 1.5k questions). Our research showcases the
importance of interpreting data in context (Ziai & Meurers, 2014;
De Kuthy et al.,
2015, 2016a, b; Ziai
et al.,
2016). The CoaLLA
project explores the integration of top-down and bottom-up
information. With Katrin
Wisniewski we explored linguistic correlates of the CEFR as part of
the MERLIN project.
As EFCamDat consultants, we collaborate
with Dora
Alexopoulou (Cambridge)
and Marije Michel
(Utrecht) to jointly analyze this very large English learner corpus
(1.18 million writing tasks by 175k learners, CEFR A1–C2). We
characterize language development both for specific constructions,
e.g., relative clauses
(Alexopoulou,
Geertzen, Korhonen & Meurers, 2015) and in terms of linguistic
complexity, emphasizing the need to account for task effects
(Alexopoulou, Michel,
Murakami & Meurers, 2017).
We also analyze L1 transfer
effects using machine learning for Native Language Identification as
an experimental testbed integrating shallow and deeper linguistic
characteristics of learner data
(Bykh & Meurers
2012, 2014, 2016; Meurers,
Krivanek & Bykh
2014; Bykh, Vajjala,
Krivanek & Meurers 2013).
Interactivity: Intelligent Tutoring Systems
Feedback
is known to be very effective in fostering learning — yet human
tutors are not always around, and the different amount of support
students get at home is a major cause for inequality in
education. While tutoring systems are increasingly taken hold in
formal domains such as mathematics and the natural sciences, foreign
language learning poses additional modeling challenges. We are
combining NLP methods with SLA insights in designing foreign language
tutoring systems that provide individual, scaffolding feedback to
students while they work on homework. Students are stepwise led to
successfully complete an exercise so that teachers in class can work
with a more homogeneous student group in class. Following the
Portuguese Intelligent Tutoring System (ITS) TAGARELA
(Amaral & Meurers 2011)
designed to complement university instruction, in collaboration with a
German school book publisher we created
the FeedBook, an interactive
workbook for English 7th grade in a DFG-funded transfer project.
In the first randomized controlled field study with an ITS fully embedded in a regular German school context,
we established the effectiveness of the specific scaffolding feedback (Meurers et al. 2019).
The approach is extended to adaptively sequence activities in the DigBinDiff project. In Interact4School, we extend the ITS with motivational feedback and explore the interface between individual learning using an ITS and the teacher orchestrated learning in the classroom. In IL2 we work on improving that interface, both on the technical side (teacher dashboards) and in terms of developing teacher training components linking SLA concepts and research results with the use of digital tools based on this foundation.
In the new BMBF project AISLA, we develop an intelligent dialog system supporting the acquisition of English in authentic, spoken language contexts.
We also collaborated
with the Tübinger Institut für
Lerntherapie in developing Prosodiya, a mobile
serious game for German dyslexic primary-school children currently
being evaluated in a large field
study with a waiting control group design.
We are developing linguistic complexity analyzers integrating a wide
range of linguistic, psycholinguistic, and SLA complexity features for
English (Vajjala & Meurers
12, 13, 14a, b, c, Chen
& Meurers 2016a, b)
and German (Hancke,
Vajjala, Meurers 12; Hancke
& Meurers 2013) — and tools such
as CTAP making it easy to use these
measures.
Applying these methods to education, we investigate
the (in)appropriateness of textbooks for students of different grades
and school types (Bryant et
al. 2017, Berendes et al., in press).
To support teachers and learners in identifying texts that are
both interesting and richly represent the language constructs to be
acquired, we created the linguistically-aware search engine FLAIR
(Chinkina & Meurers 16). On this
basis, we collaborate in the BMBF-funded KANSAS project
with the German Institute for Adult Education (DIE)
and the Mercator Institute for Literacy and Language Education
to build a tool supporting teachers of functional literacy courses.
Connecting foundational and applied issues, we are spelling out
Krashen's i+1 input fostering learning in terms of linguistic
complexity using SyB (Chen & Meurers 17), a
syntactic benchmarking tool, and we investigate the impact of challenging
learners with such input.