====== Distributional Semantic Models (ESSLLI 2009)  ======

[[course:esslli2009:start|Start page]] –
**Schedule** –
[[course:material|Software & data sets]] –
[[course:bibliography|Bibliography]]


===== Schedule & handouts =====

Credits: Some of the slides in the course are based on an earlier DSM course taught by Marco Baroni

== Introduction ==
{{esslli.dsm.intro.pdf|Slides}}

== Foundations of DSMs (Mon, 27 July) ==

  * general introduction & background
  * DSMs in a nutshell
  * the DSM parameter space: a taxonomy
{{esslli.dsm.1.pdf|Slides}}, {{esslli.dsm.1.handout.pdf | Handout}} 
== Matrix algebra & vector spaces (Tue, 28 July) ==

  * matrix representation, foundations of matrix algebra
  * geometric analysis (norms, distances, similarity & angles)
  * implementation: R as an interactive DSM laboratory
{{02_matrix_algebra.slides.pdf|Slides}}, {{02_matrix_algebra.handout.pdf|Handout}} – {{dsm_lab.R|DSM lab (R script)}}, [[http://www.collocations.de/data/bnc_vobj_filtered.txt.gz|bnc_vobj_filtered.txt.gz]] (15 MB) – {{homework.r|Homework (R script)}}, [[http://www.collocations.de/data/potter_tokens.txt.gz|potter_tokens.txt.gz]] (8.9M)

== DSMs in computational linguistics and cognitive science (Wed, 29 July) ==

  * What can we do with DSM vectors?
  * Do DSMs capture interesting aspects of word meaning?
{{esslli.dsm.2.pdf|Slides}}, {{esslli.dsm.2.handout.pdf | Handout}}, {{bnc.svd.concnouns.hclus.pdf | Concrete nouns clustering}}, {{bnc.svd.concnouns.mds.pdf | Concrete nouns MDS}}

== Geometric interpretation of DSM spaces (Thu, 30 July) ==

  * more advanced matrix algebra
  * distances, norms and angles
  * principal components analysis (PCA) as noise reduction
{{04_geometric_methods.slides.pdf|Slides}}, {{04_geometric_methods.handout.pdf|Handout}}

== SVD (Fri, 31 July, part A) ==

  * singular value decomposition (SVD)
  * managing high-dimensional DSMs (SVD, random indexing)
{{05a_svd.slides.pdf|Slides}}, {{05a_svd.handout.pdf|Handout}}

== Challenges for DSMs & current research (Fri, 31 July, part B) ==

  * looking for a unified model
  * DSMs and compositionality

{{esslli.dsm.3.pdf|Slides}}, {{esslli.dsm.3.handout.pdf | Handout}}