Near Neighbors

This program allows you to submit a term or a short text and receive a list of terms nearest the submission (in an LSA semantic space). For example, submitting the term Freud and using the psychology semantic space might return:

 

The 5 terms in the psychology space most similar to the submitted document are:
LSA
Similarity
Term
0.97 freud
0.80 sigmund
0.58 trans
0.58 strachey
0.55 unconscious

Return to Home

 

 

 


Matrix Comparison

This program allows you to submit n terms/texts and receive an n x n matrix of the cosine comparisons between the texts. Given the terms mouse cat dog house, and using the encyclopedia space this program would return:

The submitted texts' similarity matrix (in term space):
Document mouse cat dog house
mouse 1 0.42 0.14 0.05
cat 0.42 1 0.19 0.05
dog 0.14 0.19 1 0.02
house 0.05 0.05 0.02 1

Return to Home

 

 

 


Sentence to Sentence Comparison

This program allows you to submit n sentences and receive n-1 lines of the cosine comparisons between the sentences. Given the sentences This is the first sentence. Please type the sentence.Welcome to the LSA website. Here is the last one., and using the General_Knowledge_Adult(300 factors)space this program would return:

The submitted texts' sentence to sentence coherence:

COS SENTENCES
0.91
0.02
0.14
1: This is the first sentence.
2: Please type the sentence.
3: Welcome to the LSA website.
4: Here is the last one.

Mean of the Sentence to Sentence Coherence is: 0.35

Standard deviation of the Sentence to Sentence is: 0.39

Return to Home

 

 

 


One-To-Many Comparison

This program allows you to submit 1 primary text and n other texts to compare to it. You will receive a 1 x n matrix of the cosine comparisons. For example, with mouse as the primary text and with cat dog house as the other texts, (with the vector length option turned on) the program will return:

The submitted texts' similarity matrix (in term to term space):
Texts mouse
cat 0.42
dog 0.14
house 0.05

Texts Vector Length
mouse 0.33
cat 0.74
dog 2.02
house 1.72

Return to Home

 

 

 


Pairwise Comparison

This program permits you to enter any (even) number of texts and receive a similarity comparison for each pair. So for the example input mouse cat dog house, the program might return:

The submitted texts' similarity (in document to document space):
Texts cat
mouse 0.34

Texts house
dog 0.02

Return to Home

 

 

 

 

 

 


Educational Text Selection

This application allows you to explore the use of LSA as a tool for selecting texts that will augment learning. The demonstration application shows how LSA might be used to select a text about the heart based on the knowledge demonstrated in a short essay. The returned text should be understandable to the reader as well as help him or her learn something new.

 

 

 

Return to Home

 

 

 


Cross Language Retrieval

This application allows you to enter a query in one language and receive appropriate texts in another.

 

 

Return to Home

 

 

 


Essay Scoring

This application of LSA allows you to explore how LSA can be used to grade essays about a given topic. This demo allows you to enter an essay on any of seven different topics and have it graded.

Associated Links

 NMSU Scoring Demo

CU Press Release 

Media Coverage

 

 

 

Return to Home

 

 

 


Summary Scoring & Revision

This application is being developed to help students learn the techniques of summarization. This link takes you to a new application, "Summary Street", which is currently being tested in middle schools. The current version uses Java and there seems to be a problem with it and Netscape Navigator. To use this application, you must use Internet Explorer as your browser.

 

LSA-Summarization Tool
Project Description by Eileen Kintsch

Why summarization? Why did we choose to support this task?

Summarization helps comprehension by focussing students' attention on the important information in what they are reading, it forces them to identify the main points and leave out the details.

Summarizing imposes a conflict between the need to cover all the topics adequately while still being concise. Balancing these conflicting demands forces students to engage in more thoughtful, deeper processing. For example, they realize that merely selecting and deleting sentences from the original text won't do the job, that they have to find ways to generalize across details and to combine several ideas in a single sentence.

This is a difficult challenge for students and engages them in more thoughtful, deeper processing than other learning strategies, such as rereading the text, circling or underlining key concepts, or phrases, or answering multiple choice questions. Because summarizing forces students to use concepts in context, they gain a better grasp of how concepts are related to each other and to what they already know.

So summarizing is a good way to really understand and learn about new and unfamiliar subject matter. Although it offers students a valuable means of extending their knowledge, like other extended writing assignments, it plays a relatively small role in instruction in public schools. Teachers simply do not have the time to read and give feedback on all of students' writing. And often it is the case that the feedback arrives too late to be of any real value.

This is where a tool that provides automatic feedback on how to improve a summary can really help. Moreover, the kind of feedback supplied by LSA is based on how words are being used in context, not just the presence or absence of a set of key words. By using the feedback, students can write and revise over and over, until they are satisfied that their summary covers the material in the source text. They can then show it to their teacher for final evaluation. In this way, the system not only helps students better understand and learn new content, it also allows them to assume much more responsibility for their learning.

What evidence is there that LSA-guided summarizing actually does improve students' understanding? For example, couldn't students write summaries that adequately cover the content by just copying a lot of key terms or copying whole sentences from the original text into their summaries ­ that is, activities that can be performed with little understanding? There are two answers to this question: (1) It does take some understanding to select the right words and phrases to copy; one must have some notion of what is important, though admittedly this requires less thought and effort than writing things in your own words and probably results in rather shallow learning. (2) If teachers request it, plagiarism and other kinds of checks can be installed to detect writing that is too close to the text, or that is odd in other ways, making it difficult for students to fool the system.

The summarization tool has been evaluated and modified as a result of empirical tests carried out in the classroom during the past year. The results indicate a high level of satisfaction among both teachers and students: Most students state that they would like to use it to improve the content of their writing before submitting it to their teachers for grading. The tool is still under development, however, with plans to further refine the way feedback is delivered.

Two 6th-grade teachers at Platt Middle School, Cindy Matthews and Ronald Lamb, had students use the system in order to summarize texts on the sources of energy in September, 1998. They report that in follow-up classroom discussion students displayed a deeper grasp of the topics they had summarized than of those they had merely read about. They had gained "ownership" of these topics which allowed them to contribute genuinely pertinent arguments and details to the discussion. Although, we cannot claim to have proven an advantage of LSA-guided summarization over traditional summary writing, we observe that students concentrate well on their writing when they can get rapid and specific feedback and that they stay focused on the task for extended periods of time. An encouraging result of the recent trial is that two-thirds of the students were able to improve their initial summaries of the energy texts in terms of content coverage and length constraints.

Return to Home