QuizMaster

The aim of this research is to construct methods for the automatic construction of multiple-choice questions, to be used as reading comprehension quizzes.

Generating multiple-choice reading comprehension questions from a given text requires a number of steps: (1) Finding the structure in the text on which to base the question (e.g. a sentence); (2) Finding the substructure over which to abstract (e.g. a nominal phrase in the sentence); (3) Generating the surface form of the question (i.e. exactly how should the question be enunciated?); (4) Finding so-called "distractors" (incorrect but plausible alternative answers).

As a very simple example, consider the sentence "Edison invented the phonograph." This can be abstracted, e.g. into "X invented the phonograph", which can then be transformed into the question "Who invented the phonograph?" (requiring the system to realise that Edison is a person).

The four basic steps listed above can be elaborated in order to construct more complex questions by combining information from several sentences. This requires the system to resolve references, inferring relations between sentences, and similar non-local discourse phenomena. Finding suitable distractors requires both syntactic and semantic reasoning. In the project, we study and develop methods for solving these intricate problems.

The project is a collaboration between KTH and Lingio AB.

Staff:

Funding: Vinnova

Runs until 2022-11-01