Metatagger "results" inserting bad stuff! help needed

System

Ok i have tried to index all of our web sites keywords and page descriptions as both text files and as their native word docs - but it appears the suggestions arent looking at the files we have trained, and directly stripping content from the DCRs - and inserting some garbage which ive italicized

Keywords: college, education, students, mackie, brown, college education, brown mackie college location, financial assistance, globalglobalaboutusnullbrown mackie college, working life

Description: globalglobalaboutusnullbrown mackie college: about usabout usthe brown mackie college system of schools is dedicated to helping our students successfully pursue a college education that will prepare them to launch a career or enhance their current one in an increasingly competitive economy., but with the unique course delivery format offered by the brown mackie college system of schools, students are able to find a balance between the demands of school and life., students choose from programs that will enable them to become part of the growing fields of business, health care, legal assistance, and computer technology.

------------ CONFIG --------------

Description Delete Rack Script: parse -useLexicon edmc_brownmackie_content_index5 -mode sentence ;
summarize -db edmc_brownmackie_content_index5 -keysentences 3 Field Type: Summarization based Keywords Delete Rack Script: parse -useLexicon edmc_brownmackie_content_index5 -mode phrase ;
summarize -db edmc_brownmackie_content_index5 -keywords 5 -keyphrases 5 Field Type: Summarization based

Migrateduser

From the way the string is being returned with spaces removed, I would guess that you are running 'iwmthtmlproc' in a pre-processor stage, which just strips out all HTML Tags and extract the element text and that is what the summarizer would see.
ex:

snkmoorthy
too old
nothing

This would get processed by iwmthtmlproc as:

snkmoorthytoo oldnothing

probably what you need is a more intelligent pre-processor which can output, in this case, well-spaced words. I think there is an XSL solution in one of the MT plug-in processor webcasts.