I've done a lot of work with one such organization in Canada and every field worker I've talked to has had the same problem. To understand it, bear in mind that files last 20 years or more, burnout rates among staff are high, and the highest risk caregivers are also the ones most likely to move, or have people move in with them, without notice.
What happens more often than not is that what should be a safe home for the child, and therefore for the child care worker, becomes dangerous because someone not mentioned in the case worker's file either enters or leaves the caregiver's life. In the movie stereotype, this is usually a hulking ex-boyfriend out on parole, but in real life good people forced to deal with relationship changes also fall prey to cultural traditions, drugs, alcoholism, religion, or simply become unable to function for reasons we can't readily identify but might as well classify as accumulated stress fatigue.
No matter what the cause, however, family change occurs and can have negative consequences for the caregiver, the child, and the child care worker. As a result the first and most important concern in any caseworker's mind as she plans her "surprise" home visits is about whether or not she needs to request a police escort.
One of the odd things about the operation of the social welfare system is that somebody, somewhere, in at least one arm of the total system - justice, welfare, immigration, mental health services - somebody knows what the social worker needs to know. Did Joe, a violent career criminal from another city, announce his intention of living with his sister--and twin three-year-olds--to an outreach worker at the John Howard Society's halfway house? Somebody somewhere in the system knows, but the case worker doesn't.
Thus, information systems supporting child welfare workers should focus entirely on bringing the information held by those who know, to the attention of those who need to know -everything else, from compliance management to payment tracking, should be secondary and could be handled by clerks if need be.
Unfortunately, the data processing people who usually get charge of the design phase know how to do the secondary stuff, but not the primary stuff. As a result, every system I've seen (I've reviewed three in detail, another half dozen superficially) has been like the story of the drunk with the lost car keys. You know, the one who loses them in a dark area but searches for them under the streetlight, because that's where the light is? That's what these designs do -- and I'm pretty sure what the FBI virtual case file write-off both did and will do- they search where the light is, not where the keys are.
I've barely heard of "computational linguistics," and certainly not studied it, but it's a discipline that clearly has some of the right answers for this application. In practice, computational linguistics covers the art of using computers to link documents related by their information content rather than merely by the words they contain.
BRS search under SunOS 4.1 in 1988 offered both a free text search capability like Google's and a controlled vocabulary search like Yahoo's that required someone to tag the documents according to key index terms before storage. Both could, furthermore, be extended through the thesaurus facility -- thus a search for "Unix" would now hit documents containing "Linux", "BSD", and "Solaris" too.
What it couldn't do was develop and adapt its own thesaurus and document tags as new documents were added, map document connections to show holes in the data, find and prove connections where none were obvious before, or apply what amounts to a thesaurus for phrases rather than words to improve generic search quality. Some modern tools, however, can -- and should therefore be directly applicable to the problem.
Imagine a process under which everyone in the system dictates a few minutes of daily notes every day -focusing on the "he said, she said" of daily interaction with "clients" and contacts. Put this stuff in a central repository, process it using a "computational linguistics engine," and then give everyone access to it. With the right software, the caseworker would know what the outreach worker knew -- and she would be better off, the twins would be better off, and the overall system would work better.
Such software either exists now or soon will. For example, work shown off by Steve Green and some colleagues at Sun's recent laboratory open house looks like a very good fit for much of this --equally applicable whether the problem is cast as a child welfare worker trying to decide whether to ask for a police escort, or as an FBI field agent trying to clear up a hunch about a possible criminal or terrorist suspect.