gear iconCognitive Systems for Automated Story Comprehension

Story comprehension and more generally text comprehension rests on integrating the explicit information given in the text with common sense knowledge about our physical and mental worlds. There are several important issues that enter into forming a successful such integration, that have and are being extensively studied in various areas of Cognitive Psychology. These problems include the recognition of context or contexts in the text or story, the coherence of comprehension for a well-connected understanding of the text or story and the possible revision of our understanding as the text or story unfolds.

However, in trying to automate comprehension it is first necessary to be able to set up an interface between explicit text information and implicit background knowledge. This requires an automatic computational process for extracting and linking the explicit text narrative with the knowledge structures through which the common sense knowledge is acquired and stored in our computer systems.

This hackathon topic challenges the participants to explore the capabilities of today's tools for Natural Language Processing and Story Comprehension to set up such an automatic link between the text of a story in its natural language form and the relevant common sense background knowledge.

The general task is to extract the explicit narrative conveyed in the story text in a logical form that can automatically interface with the STAR system for Story Comprehension through Argumentation (

Specifically, the task contains three points of challenge:

1) Extract the logical relations and facts conveyed in the text linking correctly the relations and the objects having the relations.

2) Extract the relative time relations between the logical facts formed in (1).

3) Ensure a level of "syntactic plasticity'' so that the logical facts extracted in (1) can match the logical language used in the Common Sense Knowledge.


Here is a simple example illustrating what is required. Given the simple story:

Mary was sleeping.

Her phone rang.

She was annoyed.

Mary answered the phone.

Ann told her the good news.


the following captures the narrative in a form suitable for the STAR system:

s(1) :: sleep(mary) at 1.

s(1) :: ring(phone) at 2.

s(1) :: has(mary,phone) at 2.

s(1) :: annoyed(mary) at 3.


s(2) :: answer(mary,phone) at 4.

s(2) :: say(ann,mary,good_news) at 5.


The STAR system with its relevant background knowledge will then be able to answer comprehension questions such as:

q(1) ?? sleep(mary) at 3.

q(2) ?? ringing(phone) at 4.

q(3) ?? annoyed(mary) at 4.

q(4) ?? annoyed(mary) at 6. OR q(4) ?? happy(mary) at 6.


For the particular hackathon task challenge we will assume that all stories will be within a pre-defined theme (or domain of discourse) whose vocabulary will be given a-priori. This vocabulary will also be the vocabulary of the common sense knowledge for the theme of the stories. The background common sense knowledge for the theme will be provided although participants may wish to alter it or to provide a further structure to link to it when addressing the third challenge point of syntactic plasticity (point (3) above).


To help prepare for this hackathon task participants can familiarize themselves with the following tools:

1) STAR: Story Comprehension through Argumentation ( [STAR slides][Example stories & instructions].

2) NLP tools such as Stanford's CoreNLP or Google's SyntaxNet

3) Time extraction tools such as CAEVO (CAscading EVent Ordering system), ClearTK, TARSQI.