Topics in Cognitive Modelling

Reading List 2015

Note that this list is subject to minor changes.

Students will not be expected to read every paper on the reading list; instead, they will choose papers they are more interested in (with some constraints). Students may also propose their own readings, to be approved by the instructor. We will give more details on this during the first lecture but essentially each student will be assigned a topic to present and, together with one or two other students, will be responsible for a single class on that topic in which the group will present two modelling papers. Some topics have enough papers that we may use two classes (and two groups of students) to cover that topic. For each presentation day, all students who are not presenting will be expected to read and respond briefly to one modelling paper listed under that day's topic.

Accessing readings using links on this page: Some links on this page require a username and password (this is to respect copyright laws, which allow distribution of certain materials for course use only, or because the articles are not yet in final versions). We will distribute the password information in class. You may also find that some links only work from computers inside the University or for some reason don't work; in these cases you should still be able to find the articles through the library (see below).

Accessing readings without links: These readings should all be available through the library's e-journal services . You will need to log in through Ease in order to access journals online. You can either look up the journal by name, and then find the volume and issue you need, or (in many cases) if you are inside the University's intranet, you can search for the article title in Google Scholar and will see a 'findit@:edinburgh' link. Clicking on this link should take you straight to a page with a link to the article.

Background reading

Fundamental debates:

Are humans born with strong domain-specific knowledge and learning abilities, or is most of what we know learned through experience, using powerful domain-general learning mechanisms? Are mental representations symbolic or distributed? How is cognition affected by the nature of our physical being and its interaction with the world?

Modelling approaches:

Technical tutorials:

None of these is required reading, but they may be useful for those with less background in these areas in order to better understand some of the required readings.

Topical reading

Papers are divided into topics, with 2 or more modelling papers per topic, each of which is marked according to the modelling approach taken:

In addition, there are [Emp] (empirical) and [Rev] (review) articles given for many topics. These are not required reading but may provide useful context, especially for students preparing presentations on that topic or those who want to explore the topic further out of interest or for their final paper.

[Emp] and [Rev] papers may not be selected as your one required reading per topic.

Segmentation of sequences and scenes:

The work in this area often goes by the term "statistical learning" and much of it has been done using language-like stimuli, but there are several papers suggesting it is a more domain-general learning mechanism. We have included both linguistic and non-linguistic papers/models below.

Learning past tense verbs and the rules vs. analogies debate:

The past tense of English verbs has received a lot of attention due to the presence of both regular (walk-walked) and irregular (run-ran) verbs. The regular-irregular distinction shows up in lots of other areas of language but verbs have been used as a case example. The question under debate is how these verbs are processed: is there one system (analogy) that works for both, or two systems (rules plus exceptions)? How could these be modeled?

Grammar (learning and knowledge):

Syntax appears to be highly structured and governed by rules, at least according to most linguistic theories. Is it possible to get this behavior using distributed representations or other non-rule-based models? Can we use computational models to help us understand what kinds of grammars children actually have?

Categorization by adults:

When people encounter an object they have never seen before, they make generalizations about its behavior and talk about the category it belongs to (e.g. this furry thing I see is a cat). How do we do this, and what is our mental representation of categories?

Categorization and development:

How do infants learn to categorize objects? What role does perception play, and are there other factors at work? How is it the same/different than adult categorization?

Semantic representations:

How do people learn and represent the meanings of words and concepts?

Development of object knowledge:

As adults, we know that objects are permanent (do not disappear unless under some outside force), move as one thing, and have many other physical properties. How does this knowledge develop in infants, and is it learned?

Learning about causal relationships:

How do we infer from observations that one thing causes another? Sometimes we do this based on only one or two observations. How and in what circumstances?

Abstraction, overhypotheses, and the shape bias:

An "overhypothesis" is a fancy word for a hypothesis about hypotheses, or (in Bayesian terminology) a hierarchical model. This concept has been used to explain how we might generalize correctly to a new situation when very little data is observed for that situation; for example, if we have seen several bags of marbles containing all green, all blue, or all red marbles, and we pull a single black marble from a new bag, we might predict with high certainty that the next marble will be black, because we've learned an overhypothesis that all bags contain only one color of marbles.

Reasoning:

Visual attention:

Deciding between options:

How do people decide which option to choose from a set of possibilities? When deciding repeatedly, how do people balancing exploring -- learning which options are good by trying them out -- and exploiting the option that currently seems to be the best? How do we explain Hick's law, a systematic relationship between the number of options and the time it takes people to decide?

Sensorimotor learning and control:

How do people integrate their perceptions of the world to formulate movements and motor plans?


Home : Teaching : Courses : Tcm 

Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB, Scotland, UK
Tel: +44 131 651 5661, Fax: +44 131 651 1426, E-mail: school-office@inf.ed.ac.uk
Please contact our webadmin with any comments or corrections. Logging and Cookies
Unless explicitly stated otherwise, all material is copyright © The University of Edinburgh