Searching information in a collection of video-lectures

Searching information in a collection of video-lectures Marco Ronchetti Giuseppe Riccardi Angela Fogarolli Dipartim. Informatica e Telecomunicazioni Universit à di Trento Italy

What do we need e-learning for?

Help traditional students

e.g. provide learning material on-line

Support remote students

Make some activity faster

e.g. semi-automatically grading tests

Change the way we teach

Make it more collaborative

Create a business

Get new markets (e.g. Athabasca)

Sell content (e.g. SCORM lectures)

What do we need e-learning for?

Help traditional students

Support remote students

Make some activity faster

e.g. semi-automatically grading tests

Change the way we teach

Make it more collaborative

Create a business

Get new markets (e.g. Athabasca)

Sell content (e.g. SCORM lectures)

Traditional and remote students: how?

Traditional lecturing is dead.

The lecture model is under attack by people proposing learner centered learning

E-learning is an opportunity for a change

However…

Traditional and remote students: how?

… what is going on in your classroom?

At our place, we sometimes use a blended approach, much stuff is available through the net,

but the classroom is still there…

Can we make it any better, while the old paradigm still survives?

Breaking the space-temporal unit

Can we free the student from

spatial and temporal constraints?

Classical solutions:

Paper-based learning material

VHS cassettes (and TV or satellite broadcasting)

Video distribution through the Internet was not feasible (at least in Europe) until recently

Video is costly . Video is boring!

Alternative: podcast - vodcast

Logical Architecture Synchronous ( webcast ) Asynchronous ( On line , or download - podcast ) Asynchronous( CD - DVD )

Our requirements

the lectures should be easily browsed , with some form of indexing and a direct access to any time-location in the lecture;

lectures should be visible on all major platforms (Windows/Linux/Macintosh)

lectures should be available (in some form) also to students who do not have a large band Internet connection ;

Our requirements

impact on lectures should be minimal;

production costs should be minimal , so as to allow scaling the approach to most courses.

Set-up must be lightweight!

Operations must be simple (low level skills!)

Post-processing operations must be light (i.e. almost automatic), and not require much work

System Architecture OnStage Web Server Video Streamer DVD Browser LODE acquisition

End user requirements

Windows:

Internet Explorer or Mozilla Firefox

Java and Javascript must be enabled

QuickTime plug in

Mac OsX:

Mozilla Firefox

Java and Javascript must be enabled

QuickTime plug in

Linux:

No QuickTime plug in available!

Partial support

Cognitive interface

The main focus is on the projected slide ,

a clear voice is very important,

the video carries additional information like gestures , and can show the environment just enough

Navigation is possible

It must be also possible to use the blackboard!

LODE: the user interface

New Flash version

Lode - portability Supporto della didattica tradizionale Mac – Win - Linux Player MP4 (no slides yet) Presently attempting a port to mobile phones

Acquisition - User interface

Acquisition setup

Our experience

Over three years: we recorded

Eight 50 hours courses for the basic level

Large number of students (over 100)

Two 50 hours courses for the master level

International students (Erasmus Mundus Program)

Six 20 hours international summer schools and workshops held in Italy and Spain

We will use the system for a EASTWEB Autumn School in Korea in November 2007

Student feedback

Enthusiastic .

We were “forced” to extend the experiment to a second course!

75% anticipated using the system often or very often 5% 63% total 2 nd exam (23 stud.) 1 st exam (45 stud.) students who… 4% 6% … followed the course completely off-line 87% 51% … used the system to review at least one lecture

Advantages from student’s perspective

Better time management:

ability to recover lectures lost due to forced absence (illness, work or other time-frame incompatibility);

ability to better organize their time , deciding not to be present at some lecture (elective absence);

Better understanding

review some critical point (cases of poor understanding of a section due to concentration drop, excessive speed in an explanation or intrinsic difficulty);

ability to check the correctness of notes taken during a lecture;

Other advantages

Miscellaneous

perception by the student of a better service provided by the university ;

support foreign (italian) students who might have difficulties with the Italian (English) language ;

enrichment of the e-learning portfolio;

possibility for the teacher to view himself;

possibility to reuse lectures (across time, or in different contexts!)

Can we do more?

Scenario:

Thousands of recorded lectures.

Can we dig in them to extract material for (less formal) learning?

Cfr. approach by Mike Wald

(subtitling for accessibility)

Needle - search results

Needle - architecture ASR Multimodal search!

Needle - ASR

We used 2 ASR:

A) a speaker independent research system, used for transcriptions of the Italian parliament

B) a commercial (dictation) system

A) gave 60% word accuracy on sample

B) gave 76,6% word accuracy (after speaker training, in simulated environment)

Why so bad? Mike, but not only…

Needle - Vocabulary Research system vocabulary contained 62,879 words Manual transcription contained 702 terms - 57 not present in vocabulary 36 were technical terms Rest was english or spanish words, company names, other minor stuff (re-…, diminutives). Language model is important! We took Java books (11,443 terms before stemming - added to the vocabulary and decreased the unrecognized words from 8% to 3%)

ASR - is it good? For information retrieval - yes. Rare false positive on interesting domain!

Conclusion