Fwd: [CLS] CLS: Learning Compositional Semantics

4 views
Skip to first unread message

Jelle Zuidema

unread,
Apr 25, 2012, 9:27:32 AM4/25/12
to complin...@googlegroups.com



Sent from phone


-------- Original message --------
Subject: [CLS] CLS: Learning Compositional Semantics
From: "Zuidema, Jelle" <W.H.Z...@uva.nl>
To: "complin...@googlegroups.com" <complin...@googlegroups.com>
CC: "Latour, Sander" <M.S.L...@uva.nl>,"Smit, Frank" <F.S...@uva.nl>


Dear all,

Next week, we'll have (finally) a CLS meeting again - and we'll make it a double bill! The theme is Learning Compositional Semantics, but the to talks approach it from quite different directions.

Sander Latour and Frank Smit will present their game-like environment used to generate a semantically-annotated corpus that might be use for learning. Phong Le will review the recent literature on machine learning of semantic parsers - many of them learning from semantically annotated corpora.

Date: April 25th
Location: ScP B0.209

Program:
16h00-16h45 Sander Latour & Frank Smit, Semantics Construction Environment for Non-Experts (abstract below)
(including 15 mins discussion)
16h45-17h15 Phong Le: Learning Semantic Parsing

***

Not really CLS, but interesting for many of you, so don't forget Tecumseh Fitch's SMART Cognitive Science Lecture this Friday, at 4pm in UT3.01 (Nieuwe Doelenstraat 16). See http://smartcognitivescience.wordpress.com

***

Semantics Construction Environment for Non-Experts
Sander Latour & Frank Smit

Abstract:
Over a one month period we've developed a series of online games that require the user to convert pictorial scenes to natural language sentences and vice versa. The result of these games was a parallel corpus of english sentences aligned with a structural description of the scenes. These structural descriptions cater for conversion to a desired formal semantic representation. The games were played by non-expert users in the Amazon Mechanical Turk environment, of which the corpus was evaluated manually on usefulness. The result of this project is a proof of concept that shows the potential of the game-based approach in creating corpora.


Reply all
Reply to author
Forward
0 new messages