Webcrow Challenge: An AI Approch for Crossworld Puzzles

Date

Feb 24, 2023, 10:00 am to 12:00 am

Objective (one research question): 

WebCrow is an AI software developed to solve multilingual crossword puzzles using NLP technology [1]. NLP approaches may exploit different sources of knowledge like ontologies, knowledge graphs and the web. Then the system attempts to provide the correct answer for each textual input clue, by means of a sequence of logic and linguistic steps.

The focus of the workshop will be on highlighting some research lines connecting neural-symbolic approaches and crossword puzzle solvers and generators. In particular, we aim at presenting WebCrow as a candidate for defining a clear benchmark for neural-symbolic systems.

Organizing partner: 

CINI, Expert.AI

Participating partners: 

...

WP 4 Task: 

Task 4.3

Expected output: 

• Discussion on new ideas for improving the Webcrow solver performances by exploiting ontologies, especially in multi-language frameworks. 

• Promising directions for the construction and evaluation of new benchmarks for neural-symbolic AI on crossword puzzles.

• Setting up new collaborations (e.g. a new Coordinate Action) to work on open research directions.

Bibliography: 

[1] Zugarini, Andrea, and Marco Ernandes. "A Multi-Strategy Approach to Crossword Clue Answer Retrieval and Ranking." CLiC-it. 2021.

[2] Marco Ernandes, Giovanni Angelini, and Marco Gori. 2008. A web-based agent challenges human experts on crosswords. AI Magazine, 29(1):77–77.

[3] Ernandes, Marco, Giovanni Angelini, and Marco Gori. "Webcrow: A web-based system for crossword solving." AAAI. 2005.