Linguistic Modeling and its Interfaces
Oberseminar, Detmar Meurers, Winter Semester 2015
This series features presentations and discussions of current issues in linguistic modeling and its interfaces. This includes linguistic modeling in computational linguistics, language acquisition research, Intelligent Computer-Assisted Language Learning, and education, as well as theoretical linguistic research with a focus on the interfaces of syntax and information structure. It is open to anyone interested in this interdisciplinary enterprise.
A list of talks in previous semesters can be found here:
Abstracts This talk presents an end-to-end automatic question generation system for
authentic texts, using a novel approach based on customizable patterns. The goals of the
system are to ensure a text has been read and verify basic understanding of a text.
It detects the main terms in the texts, generates questions about them and selects
appropriate distractors for them — all are based purely on the given text itself. Additionally,
ranking and filtering mechanisms are implemented for choosing the best tasks. The user
can modify the task patterns and create new ones for making the system focus on
the desired contents in the text and generate questions best suited to the learning
goals.
Abstracts This thesis reports on the implementation of linguistically motivated features for the task
of readabiliy classification of German. Five feature sets were designed, containing overall 46
features that are either a) linguistically more detailed re- implementations of common readability
features, such as complex NPs and VPs, length measures of distances between Topological Field
positions or inference markers of conditional clauses; or b) new features based on recent
insights from research on the German academic writing register. These features were
incorporated to the readability classifier by Hancke (2013) and Hancke, Vajjala & Meurers
(2012) with and without the enhancements by Galasso (2014). Their performance was
tested on the Reading Demands corpus consisting of texts from textbooks from German
secondary schools by four different publishers. The classifier enhanced with the new features
achieves an accuracy of up to 53.93% for grade level and up to 76.86% for school type
classification and it improves classification compared to the non-enhanced classifier by up
to 0.52% for grade- and 1.28% for school-wise classification. Also, the new features
allow for detailed insights in the diverging linguistic properties of texts from different
publishers.
_________________________________________________________________________________
Last updated: November 24, 2016