Homework
Homework 6 (Dues Monday, April 2, by the beginning of class)
For this homework, we will explore some of the ambiguities of language that make natural language understanding difficult. The homework has two parts.
Part I: Ambiguous grammar parsings
I owe this homework to Chris Reisbeck, a well-known NLP researcher who has a hobby of collecting ambiguous headlines such as this one from Friday’s USA Today: “Volunteer rates booming.” Are the number of volunteers going up, or did a single volunteer rate the effectiveness of a boom? Many headlines, of course, provide even more humorous alternative parsings.
Find a newspaper headline that has at least two potential parsings. Give each of the grammar parsings as a parse tree, and then paraphrase the result. Here are two example parsings for a famously ambiguous sentence: “Time flies like on arrow.”
In this first parse, time is a thing that moves in the manner of an arrow (click on the thumbnail to get the full image)
:
In the second parse, “time flies” are insects that, well, enjoy a good arrow now and then:
Part II: How good is current online translation?
There is a famous apocryphal story about the early days of NLP, when a researcher system by first translating a phrase from English to Russian, and then translating the result had into English. The phrase used was” The spirit is willing but the flesh is weak.” Unfortunately, the system translated this as “The wine is good but the meat is rotten.”
Test a current online translation by going here and clicking on the “IM Translator” link. Then decide on a phrase to translate to a foreign language of your choice and then translate it back to English. Give the original phrase and both translations, and speculate on how well the translation worked and why.
Homework 5 (Due Friday, March 16, by beginning of class)
Learning Objective
In this homework, you will take part in a neurological test of face recognition and consider how the computer model of mind informs our understanding of neurological deficits.
Assignment
1. Go to this link and take both of the online recognition tests (”Famous Faces” and “Old Faces / New Faces”). Each test should take less than 20 minutes.
2. When you finish each test, print out your results to turn in. Please black out your result numbers if you do not want to include them.
3. With the results pages, please also answer the following questions:
a. What is prosopagnosia, and how does it differ from other kinds of agnosia?
b. If this test were performed using buildings instead of faces, how do you think that you would perform? How would a prosopagnosiac perform?
c. Could you account for prosopagnosia within Marr’s theory of vision. Why or why not? (There is no “correct” answer here — I’m looking for how our computer models of vision might inform our understanding of neurology).
Homework 4 (Due Monday, February 26, by beginning of class)
- How many moving objects can we track at once if all the objects are within the visual field? Try this with some real-world objects (fish in a aquarium, friends moving their hands, etc.) and try to come up with a number.
- Give your conclusion
- Explain your methodology and how it supports your conclusion.
- Look at the letter confusion matrix given here (Experiment 1 results only).
- Based on the confusion matrix and a Pandamonium-style model, what letter features might underlie the frequently-confused letters? Describe your reasoning, and list specific letters and features as examples.
The confusion matrix given here is from: Gibson, E. J., Osser, H., Schiff, W., & Smith, J. (1963). An analysis of critical features of letters, tested by a confusion matrix. Cooperative Research Project No. 639, U.S. Dept. of Education and is hosted by the Letter Similarity Data Set Archive.
Homework 3 (Due Friday, February 16 at the beginning of class)
Please complete a 3-page (or longer) essay on the following question:
Describe the multiple drafts model. Contrast it with the classical view of consciousness. Which seems more feasible and why?
Please submit the summary essay in this simple format: a 1-1/2 spaced essay using a 12-point Times Roman font and margins of 1.25” on the left and right side and 1.0” at the top and bottom (except for the spacing, this is the typical default for Microsoft Word). At the top of the page, please write a single-spaced header that contains a descriptive title (centered), your name, “Homework 3”, and your email.
For the essay itself, here are some useful heuristics: try to have a specific claim that is the point of the essay, and get to that claim early (by the end of the first paragraph, if you can manage it). Then support that claim using a logical structure where each supporting point is handled by a single paragraph where the first paragraph gives that point. Using this style can help you create a readable essay. Write for the general reader rather than a professor or fellow student.
The grade (20 points) will depend about 2/3rds on content and about 1/3 on readability and style.
Homework 2: Calculate the similarity of animal concepts using MDS and feature-based models (30 points)
Due: Wednesday February 7 at the start of class.
Motivation: By working through two examples of these models, you’ll gain a deeper insight of how well these analogy models work. You’ll also have a better idea of the power and limitations of these models.
Part 1: Multidimensional Scaling
- Pick seven animals from the set Rumelhart & Abrahamson article we discussed earlier. Specifically, seven animals from the following list: hawk, eagle, goose, duck, chicken, pigeon, parrot, parakeet, robin, sparrow, cardinal, blue jay, goat, sheep, cow, horse, pig, deer, bear, lion, dog, rabbit, cat, and mouse.
- For each pair of items from the list, giving a percentage rating of how similar the two concepts seem to you (e.g., you might rate a pen and a pencil as about .76).
- Use the PERMAP tool (installation tips and instructions here) to create a 2-D similarity space. How well does this space fit with your own intuitions about how similar the animals are?
Part 2: Tversky’s Contrast Model
- Drop two animals from your list. Then, for each of the five animals, generate a list of features for that animal. Try to be fairly complete, but don’t worry about getting every possible feature (e.g., you don’t need to include “has mass”). One heuristic that you might use is to time yourself, and attempt to do each list in about 2 minutes.
- Calculate the similarity of each pair of animals using Tversky’s contrast model: s(a,b) = qf(A^B) – af(A-B) – bf(B-A). Let q =1, a=0.2, and b=0.4. Let f be a simple count of the members in the set. This should give you a table of 25 numbers (since the values are directional), although 5 of those values should be self-comparisons.
- First, examine the set of values to determine if the comparisons are not symmetric. For our example in lecture, one would expect a pen to be more similar to chalk than chalk is to a pen. Does this show up in your results?
- Next, compare the values computed for Tversky’s Contrast Model with the distances in the MDS model. Are they generally proportional? If not, when and how do they differ?
Tips:
– For some steps it may be helpful to use a spreadsheet or other calculating tool, but it isn’t strictly necessary.
– Even if you don’t have time for the homework until later, install and test the PERMAP software at least a week before the deadline. The installation script seems to be a bit buggy.
– Check the comments for this page for additional tips and FAQs.
Your final report should be typed, explain each step above, and show your results.
Extra credit (8 points): Do the whole thing, but with seven animals for both parts. Then use the Contrast Model results (using one half of the matrix) as the data for the MDS tool. If you do this, you should be able to compare the two models visually.
Homework 1 — Folk Psychology. Due Wednesday, January 17, at the start of class
Although few of us are psychologists, almost all of us have aspects of what cognitive scientists call a folk psychology — theories about the mind. One way to begin our exploration of cognitive science is to reflect on the theories that each of us have, explicitly or implicitly, about how the mind works.
For your first assignment, pick one of the questions (broadly defined) from the Boden reading (pp. 2-3) and discuss it with two friends. Don’t worry about trying to answer the question definitively, but instead try to lay out each person’s assumptions and reasoning about the area. Write a summary of your and their opinions, carefully noting the views that you held in common and where you differed. Are your disagreements something that could be resolved by experimentation, or do they depend on differing assumptions about human nature?
The summary essay should follow this format: a descriptive title, your name, and the names of your friends. The essay itself should be about 350-450 words. The essay should be typed and neat, but otherwise the format is up to you. The grade (20 points) will depend about 2/3rds on how well you explore the question (again – you don’t have to answer it) and about 1/3 on readability.
Special note: If one of your friends is a member of this class, they should choose a different question for their homework.
Please bring your essay to class on Wednesday.
CogSci 3790: Introduction to Cognitive Science 
I take it that out “two friends” don’t have to be in this course, right? (That is, any “two friends” we have; they don’t have to go to GIT, right?)
Thanks in advance,
Jimmy M. Espana
Yes, any two friends, Tech or not Tech, in course or out. In fact, technically speaking they don’t even have to be your friends. 😉
How important is it to cap the essay at 450 words? If mine is longer, do I need to be cutting information out, or is length not a big issue in this class?
It’s okay to go over a bit. I wouldn’t go over a lot, however.
Is one question on of the small paragraphs or just one of the sentences?
When you said, “pick one of the questions (broadly defined),” does that mean you literally want just one question, or does that mean that you want one of the series of interconnected questions that she presents? I interviewed my friends using one of the small series of questions because together, they more fully explored that one idea.
I worded the homework to give you the discretion to choose to do it either way. The way that you handled it is fine.
Can the numbers shown by the Tversky’s Contrast Model be negative in some cases?
How are we supposed to graph the numbers from Tversky’s Contrast Model in the MDS tool if our feature sets aren’t all the same size? The error keeps saying that the last numbers of the diagonal aren’t the same.
Are A and B just the number of traits, or the number of traits the two have in common (or something else…)?
Courtland:
Sure, they can be negative. It’s up to you to interpret them. For instance, you might try normalizing them to help you analyze and compare.
Taneshia:
The MDS model has a number indicating similarity between two entities. The feature-by-feature comparison is already factored into that similarity value.
Lisa:
The assignment says, “Let f be a simple count of the members in the set.” “The set” refers to the set of features (i.e. traits). So f(A^B) refers to the count of how many features are the same, f(A-B) refers to the count of how many features are in A but not B, and f(B-A) refers to the count of how many features are in B but not in A.
Eh, I don’t think that was what Taneshia was asking because I think she has the same issue I had. If A, the feature set for some animal, has 10 elements, then the spot on the diagonal, of the half-grid where A that permap reads, where A is compared to A will be 10. Similarly, if B has 11 elements then the spot on the diagonal where B is compared to B will be 11. However, permap is not capable of dealing with this as it intended for use with MDS models and one of the noted flaws of MDS models is that they assume that all objects are equally like themselves. In order to make all the elements on the diagonal equal, I averaged them. I was just checking that this was an acceptable way of resolving the issue.
I’m kinda confused about what the ‘classical view of consciousness’ implies. Are we talking about the wax tablet sort of thing? Or are we talking about the other views of consciousness that came about before the Multiple Drafts model? Am i completely missing something?
(happy
By the classical view of consciousness, I mean the Cartesian Theatre view.
i would like to say that I have fairly bad pacing information given when writing papers and I’ll probably end up repeating myself a bit on the last page ( more than i’m supposed to just from summarizing)