Mechanisms of Meaning

Autumn 2010


Lecturer: Raquel Fernández (ILLC, University of Amsterdam)

Timetable: Mondays 13-15h. in G2.04 until the 18th of October. Room G2.13 from 1st Nov on.

Theories of natural language meaning often take the sentence as the basic unit of study and focus on how sentential meaning is compositionally built up from the meaning of a sentence's constituents. In this course we will look into two aspects of meaning that go beyond compositional semantics. In the first part of the course, we will zoom in to explore the meaning of words, building on "Distributional Semantic Models" which assume that the meaning of a word can be inferred from its usage, i.e. from its distribution in discourse. In the second part, we will zoom out to look into how meaning arises collaboratively in dialogue, as combinations of utterances are contributed by different interlocutors.

Evaluation: Homework exercises (25%), 1 or 2 presentations of readings per person (25%), and a final paper or project plus the presentation of that project at the end of the course (50%). See details for final papers here.

This webpage will be updated throughout the semester. Please check it regularly. Its contents are subject to change.


Date Contents Homework
6 September 2010 Slides. This first lecture has given an overview of the contents to be covered in the two parts of the course. At the end, we have started to look into the theme of the first block: how to represent word meaning. Readings for next week:
* Pustejovsky (1991)
* Lapata (2001)
13 September 2010 Slides. We reviewd axiomatic and primitive-based theories of word meaning, and then focused on the Generative Lexicon framework. This is the main reference for this lecture:
  • James Pustejovsky (1995) The Genertative Lexicon, MIT Press.

A shorter introduction to GL can be found in Pustejovsky (1991), which is required reading for this week.

Readings for next week:
* Lapata (2001)
* Kilgarriff (1997)
20 September 2010 Slides. This class was dedicated to discussing the following papers: Lapata's paper presents a computational model of some aspects of GL. Kilgarriff takes a different approach, which will serve as introduction to distributional models of word meaning. See details of Stefan Evert's talk on the right.

The slides for this lecture include details of homework for the following weeks.

HW1: Attend the talk by Stefan Evert on Distributional Semantic Models at the ILLC Computatinal Linguistics Seminar on Wed 22, 4pm, and write a summary of the talk (detailed instructions).

HW1 is due on Sept 27.

27 September 2010 CLASS CANCELLED (due to SIGDIAL and Interspeech Conferences in Japan) HW2: Semantic annotation exercise. Read the instructions carefully and download the data set and the definition of the semantic relation.

HW2 is due on Oct 4.

4 October 2010 CLASS CANCELLED (due to meeting in Utrecht). Readings for next week:
* Chapter 2 "Typicality and the Classic View of Categories" from Murphy (2002).
11 October 2010 Slides. In the first part of this lecture we discussed how to assess the reliability of linguistic annotations and calculated inter-annotator agreement for the semantic annotation exercise (HW2).

In the second part, we started to look into psychological theories of concepts and word meaning. The main reference for this topic is:

Noortje Venhuizen presented chapter 2 "Typicality and the Classic View of Categories." Here are the slides of her presentation.

Readings for next week:
* Chapter 3 "Theories"
* Chapter 11 "Word Meaning"
both of them from Murphy (2002).

There should be copies available in the MoL room. Else, ask me.

18 October 2010 We will go on discussing material from Murphy (2002).
  • Chapter 3 "Theories" will be presented by Marta Sznajder (slides).
  • Chapter 11 "Word Meaning" will be presented by Adam Pantel.

Everybody should read these two chapters beforehand and come to class ready to discuss the details.

Try to decide the topic of your final paper during the exam week. You may schedule an appointment with me to discuss this.
25 October 2010 NO CLASS (exam week) HW3 on distributional semantics is due on Wednesday Nov 10. Send me your answers via email as a PDF attachement with your name.
1 November 2010 New lecture room from now on: G2.13

Slides. We will look into Distributional Semantic Models, building on what you learned from Stefan Evert's talk. We'll discuss the historical roots of these models and the core theoretical assumptions behind them. The main reference for this lecture is:

8 November 2010 We'll discuss the following papers:
15 November 2010

Slides. This has been the last lecture dedicated to Distributional Semantic Models and to word meaning in general. After discussing HW3, we looked into recent proposals for accounting for compositionality with DSMs, focusing on adjective-noun composition. References:

Make an appointment for some time this week to speak to me about your plans for your final paper.

22 November 2010

Slides. We started to look into meaning in dialogue. The lecture covered the main ideas behind the collaborative model of Herb Clark and colleagues and gave a sketch of the interactive alignment model of Pickering and Garrod. Main references:

29 November 2010

Slides. In this session we will consider the role of dialogue interaction in language acquisition. You can find several articles related to this topic on the websites of Eve Clark and Matthew Saxton, two psycholinguists that take dialogue interaction seriously. The following monographies by these researchers are good introductions to language acquisition in general, and contain chapters dedicated to interaction:

In the second part of the class, we'll discuss the following paper:

6 December 2010

Slides. We discussed dialogue phenomena that call for incremental interpretation: turn-taking, continuous meta-communicative feedback, and split utterances. Here is the seminal paper on turn-taking within the tradition of Conversation Analysis. You can find more references on the slides.

13 December 2010 We will meet in room C3.108 (the ILLC small meeting room). This final session will be dedicated to presentations that ouline your plans for your final paper.
  • Speakers: please prepare a 15-20 min presentation including:
    1. background and motivation behind your study (why you do it)
    2. your main aim (what you plan to do), and
    3. a sketch of your approach (how you'll do it).
  • Audience: please give useful feedback to your fellow class mates. Does the plan make sense? is it reasonable? Are there aspects that are not clear? Any suggestions for improvement?
Tentative schedule (surname alphabetical order):
  • 13:00-13-20h: Holger (slides)
  • 13:20-13:40h: Irma (slides)
  • 13:40-14:00h: Marta (slides)
  • 14:00-14:05h: break
  • 14:05-14:25h: Noortje (slides)
  • 14:25-14:45h: Andreas (slides)
20 December 2010 NO CLASS (exam week)

Final papers must be submitted (by email as a PDF attachment) no later than 15th of January 2011.

You may want to have a look at these general tips on how to write a paper by Ulle Endriss.


Final Papers

During the final part of the course you are supposed to write a paper about a phenomenon or a result related to the topics covered by the course. You will have to present your ideas and plan for the paper in a talk at the end of the semester. (The paper itself can be handled in a few weeks later.) The paper can be based on an article from the literature but it should also include some original contribution: for instance, you may point to an interesting connection between different approaches, or discuss the theoretical/philosophical implications of a computational/applied approach or an empirical result from the psychological literature. You may also propose an improved (or new) formalisation, or a sketch of a computational model, of a particular linguistic or psychological phenomenon. These are just some possibilities -- I'll be happy to hear your own suggestions as well.

As a starting point, you may take the paper you presented during the course or any of the following articles. You are also very welcome to suggest a paper or topic of your own choice. In any case, let me know well in advance what your chosen topic is -- definitely no later than Nov 15.