DASP303 – Computational modeling

English only

Fall 2011

Goals and content

This course is about modeling. The emphasis is on models of machine learning, particularly on distinguishing regularity from exceptions. By letting a machine mine some data, we can find regularities in the data. These regularities may be expressed in terms of pattern associations, similarities or rules. We will mostly use language data, but the methods can be generalized to other domains.

After the course, students will be able to distinguish different modeling paradigms, choose methods depending on data, apply suitable computer tools and interpret results of modeling in small problem domains.


Teaching will be in English if foreign students are attending, otherwise in Norwegian.

The teaching includes lectures and practical computer exercises. Computers with relevant software are available to course participants. The use of the computer lab presupposes some knowledge of Linux. Hardware or software failures can be signaled to bs.uib.no. Students can also use their own laptops if they have the skills to install the necessary software .


The exam consists a term paper (semesteroppgave) of about 4000 words and a written exam. The subject for the paper can be chosen rather freely but must be approved by the teacher. Normally, it involves building and/or testing a computational model of language processing. It is obligatory to hand in one draft. The teacher will provide advice on the assignment and will provide comments on drafts.

The final term paper must be handed in on My Space (Mi side). Please consult the instructions on My Space for handing in the paper. NB. If your paper is not only a pdf file, but also has appendices consisting of data sets, programs, model output, etc. it is recommended to collect all files in one archive (zip, gz or rar). These file types should also be accepted. Long appendices do not need to be printed, only provided electronically.

An example of a possibly suitable structure of the paper is the following:

  1. Introduction: describe background, goal, motivation of the approach, hypothesis and expected outcomes.
  2. Data and method: source of the data, description of the dataset, construction process of the final data from the source including selection criteria and adaptation of file structure, range of values, etc. Suitability and possible shortcomings of the data. Description of the method and programs used.
  3. Models and their performance: description of constructed models with details of parameter settings etc. Performance of each model, analysis of errors.
  4. Evaluation: comparison of models with other models or with results from the literature, with standards, etc.
  5. Discussion and conclusion: relation between models outcomes and expected outcomes, possible explanation of divergences, limitations of the models.
  6. References: list of publications that your paper refers to.
  7. Appendices: list of appendices with their descriptions.

Some advice for writing the term papers:


There will be a written exam with several essay questions. Normally each question is answered on about half a page to one page. Answer the question directly and give examples where needed. Here are some example questions from earlier exams:

  1. How can language units be represented as activation distributed over several input or output nodes in neural networks? What is an advantage and what can a problem with such a distribution? Use examples from the reading list.
  2. How can models based on memory based learning or neural networks support or disprove a theory? Discuss with models from the reading list.
  3. In which way are computer models of language processing simplifications and abstractions of the human language capacity and human language behavior?
  4. How can we evaluate models of language processing if we do not know exactly how the human language capacity works?