About a year ago I posted some emails on var­i­ous techie UIUC news­groups about the need for DDoS-FOIA soft­ware, the right book­end to the left one of DDoS-FOIA fil­ing.  With the UIUC-FOIA office neatly boxed between these endpoints.

As you can prob­a­bly tell, I’ve spent too much of my hol­i­day break research­ing and writ­ing about UIUC — call it method, call it mad­ness, I’m used to either as a call­sign.  In any case, I’ve finally fig­ured out how to eas­ily search a sin­gle site with the google, so I’ve been search­ing the UIUC-FOIA office’s site, where they post (search­able) log files of every FOIA filed.¹

When I did this search of FOIA result let­ters using “Salaita” as the search term, I found 10 hits, cor­re­spond­ing to 10 FOIA requests up on the site as of the end of 2014:²

1. 14–549 (16 pages);
2. 14–535 (143 pages);
3. 14–538 (204 pages);
4. 14–544 (78 pages);
5. 14–565 (Request to Narrow);
6. 14–584 (0 pages);
7. 14–591 (55 pages);
8. 14–617 (179 pages);
9. 14–536 (Request to Narrow);
10. 14–652 (48 pages).
11. (All of these files pack­aged as a sin­gle .zip at UIUC-Salaita-FOIAs_ASOF_20141231 – again note that these are just the response let­ters from UIUC-FOIA, and not the actual doc­u­ments.  If you want the doc­u­ments you have to FOIA UIUC to obtain them, ref­er­enc­ing the FOIA request number.)

As I say, the web­site doesn’t pro­vide the pro­duced doc­u­ments, which total up to 723 pages.  But even as fig­ure out how to pro­cure that cor­pus of Salaita doc­u­ments,³ I thought I’d present this as an instruc­tive oppor­tu­nity to col­lab­o­rate with someone(s) out there on DDoS-FOIA soft­ware adapted for parsing/crunching this set of data.

Not just for this data, of course, but for sets of FOIA data in gen­eral, because all of these sets unques­tion­ably share some very com­mon characteristics.

 

1. My back­ground in patent search­ing software.

I come to the ques­tion of DDoS-FOIA soft­ware by way of my dal­liance over the last 10 years or with cod­ing my own patent search­ing soft­ware, which I did out of com­plete frus­tra­tion with the poor ratio­nales behind cur­rent patent-search software.

So with­out going into gory detail, “nat­ural lan­guage” analy­sis applies to both the lan­guage as, say, eng­lish gram­mar, and also to the over­all rules of con­struc­tion that a par­tic­u­lar class of cor­pus falls into — the high-level “lan­guage” of the class of cor­pus.  For exam­ple, patent doc­u­ments are highly struc­tured not sim­ply by the lan­guage they’re writ­ten in, or even by the doc­u­ment struc­ture like title + abstract + spec­i­fi­ca­tion + claims, but also by the human being that con­structs the patent, that human being largely being some­one specif­i­cally trained in a rel­a­tively stan­dard way of patent construction.

Given the above, my idea was to use met­rics for pars­ing patent doc­u­ments that were indica­tive of the high-level con­sid­er­a­tions the patent attor­ney, patent agent or expe­ri­enced drafter typ­i­cally uses in prepar­ing the doc­u­ment.  For any­one famil­iar with these tech­niques, any patent or patent appli­ca­tion has mul­ti­ple “tells” in it that speak vol­umes to you … if you know what you’re look­ing for.⁴

And I do know what I’m look­ing for.  I’m a patent attor­ney, it’s what I do.⁵

 

2. The “nat­ural lan­guage” of a sin­gle FOIA pro­duc­tion set.

So to run with the above con­cept, when you get a FOIA result set back from any par­tic­u­lar FOIA office, it’s going to have embed­ded in it all the ticks and idio­syn­crasies of that office’s pro­ce­dures and per­son­nel.  And also the the ticks and idio­syn­crasies inher­ent in FOIA productions.

Let’s exam­ine these in reverse order:

1. Ticks and idio­syn­crasies of FOIA doc­u­ments gen­er­ally.  Redac­tions are of course the sig­nal fea­ture of FOIAed doc­u­ments, so any con­tem­pla­tion of analy­sis nec­es­sar­ily begins with some method of pars­ing redac­tions.  At the low­est level such pars­ing would be to iden­tify if the reac­tion was a sin­gle word, a phrase, a sen­tence, a para­graph etc.  You can go wild with this con­cept and extend it to meta-analysis of back­fill­ing of redacted words by con­text of sur­round­ing non-redacted text.  You can add to that esti­ma­tions of con­tent based on the phys­i­cal length of the redac­tion (e.g., in cm) ver­sus the kind of font used to gen­er­ate the text.  Another hook into analy­sis would be to use the stan­dard lan­guage of emails (from, to, sub­ject, attach­ments) to do fur­ther pars­ing and analysis.
2. Ticks and idio­syn­crasies of FOIA doc­u­ments result­ing from pro­ce­dure and per­son­nel.  Every FOIA office has its own sig­na­tures for what it likes to redact, and this breaks down into indi­vid­u­als within the office as well as the office over time, depend­ing upon how sophis­ti­cated the redact­ing and how much work­flow the office expe­ri­ences.  For exam­ple, redact­ing “per­sonal infor­ma­tion” is really sim­ple when it involves merely phone num­bers and social secu­rity num­bers and addresses, but much harder when it’s infor­ma­tion about, say travel, which might seem benign but often­times can be used to recon­struct adul­ter­ous adven­tures⁶ and so on.⁷ Thus doc­u­ment recon­struc­tion may often be advanced by com­pil­ing a data­base of the ticks and foibles of each FOIA office and each FOIA offi­cer.

 

3. The “nat­ural meta-language” of an aggre­gated (DDoS-FOIA) pro­duc­tion set.

FOIAed doc­u­ment recon­struc­tion becomes even more inter­ested in the con­text of a DDoS-FOIA approach, since we’re now talk­ing about a cor­pus of sets of doc­u­ments, each set obtained by sep­a­rate FOIA, each sep­a­rate FOIA pos­si­bly parsed by a dif­fer­ent FOIA offi­cer with­out coor­di­na­tion with the other FOIA officers.

So take the 723 page Salaita doc­u­ment cor­pus that I’m still try­ing to fig­ure out how to obtain.  That cor­pus resulted from 10 sep­a­rate but likely over­lap­ping FOIAs, there­fore, data-reduction is manda­tory to analysis.

But.  Data reduc­tion needs to occur against an aware­ness of pos­si­bly only par­tially–redun­dant data — Miller’s name came up because UIUC-FOIA failed to redact it in one instance, as did Miller’s office loca­tion in North­brook or travel des­ti­na­tion of Santa Fe.

And beyond this, even seem­ingly benign non-redacted text can cast a light on other redac­tions, a sit­u­a­tion that often occurs when one redac­tor removes an entire sen­tence, and another only part of a sentence.

To sum­ma­rize: aggre­gated FOIA doc­u­ments obtained by DDoS-FOIA offer up ana­lytic pos­si­bil­i­ties that likely go beyond what’s obtain­able in a sin­gle production.

 

4. A pro­posal for DDoS-FOIA software.

Based on the above, it would be com­pletely rea­son­able to take a stan­dard NLP front-end in the lan­guage of your choice and build not just a redacted doc­u­ment proces­sor but also a proces­sor for build­ing cor­re­la­tions and extract­ing infor­ma­tion from cor­puses obtained by DDoS-FOIA.

I’ve already talked about this to a num­ber of dif­fer­ent orga­ni­za­tions, none of which seemed par­tic­u­larly inter­ested.  In fact, I wanted to join the UIUC Exec­u­tive Round­table to talk about this as well as my patent search soft­ware com­pany, but was imme­di­ately rejected as “merely a patent attor­ney” by the pow­ers that be of that appar­ently exclu­sive club.⁸

So.  Are there any peo­ples out there that would be inter­ested in this?  My con­tacts are extremely good.  And it’s a con­cept eas­ily scal­able to MMDDoS-FOIA⁹ soft­ware, where per­haps a grand uni­fied data­base of cross-the-country FOIA fil­ings could be built, ana­lyzed, and discussed.

 

---

The mar­quee image started life as an inte­rior panel of “Space Detec­tive #1,” which I believe is of 1951 vin­tage.  See Space Detec­tive #1.  Obvi­ously I took a cer­tain amount of artis­tic license with the orig­i­nal, but I’m sure Teena and space detec­tive (aka Rod Hathway/The Avenger) approve.

And if you think I’m being too acer­bic about the UIUC-FOIA office, go back through my ear­lier posts and re-read that quote from the Illi­nois Attor­ney General’s office about the UIUC-FOIA office’s ongo­ing “ille­gal” behavior.

---

 

 

1. The search string to enter into the google search box is “search_term site:http://www.foia.uillinois.edu/” (with­out the quotes), where search_term is the term you want to search for that was present in the FOIA response let­ter.  Note that the UIUC-FOIA office web­site doesn’t put up the actual FOIA results, so your search won’t iden­tify the search term if it occurs in those results. [↩]
2. Illi­nois man­dates a 5 work­ing day period between receipt of FOIA request and pro­duc­tion of doc­u­ments, but that period can be extended by 5 days by the FOIA office, and still longer if the filer is a frequent-filer (“vex­a­tious filer”) or the FOIA office uses one or more of its many tricks to delay response (e.g., ask­ing the requester to nar­row the request). Con­se­quently it’s highly pos­si­ble that there were more than 10 FOIA fil­ings in 2014 con­tain­ing the key­word “Salaita.”  These will crop up on the UIUC-FOIA log site, but only after UIUC-FOIA sends out a response let­ter. [↩]
3. Given my vexatious-filer sta­tus, UIUC-FOIA will delay pro­duc­tion of these doc­u­ments for months, if not years. [↩]
4. Note that this is a blan­ket state­ment that doesn’t take into account dif­fer­ent prac­tices in dif­fer­ent coun­tries and for dif­fer­ent types of patent doc­u­ments, e.g., pro­vi­sion­als ver­sus US non-provisional fil­ing ver­sus PCT etc. [↩]
5. Not to boast, but this isn’t what most patent attor­neys do.  I have an aca­d­e­mic bent that makes it impos­si­ble for me to use a tool with­out either a) break­ing it or b) mak­ing it bet­ter.  I’d like to add a pos­si­bil­ity of c) break­ing it in order to make it bet­ter … some­times I think I do this, but more often I prob­a­bly just break things. [↩]
6. “Hik­ing the Appalachian Trail.” [↩]
7. As I said in an ear­lier post, UIUC-FOIA redacted Steve Miller’s trip des­ti­na­tion of Santa Fe in some emails and not in oth­ers.  And didn’t redact Chan­cel­lor Wise say­ing “Oh, I love Santa Fe.”  So much for con­sis­tency. [↩]
8. Part of a drawn-out process of expe­ri­enc­ing what, in my opin­ion and frankly that of oth­ers, is a fre­quently sub-par unit of UIUC, the Research Park, or at least the incu­ba­tor com­po­nent thereof. [↩]
9. That is, “Massively-Mutlple Distributed-Denial-of-Stonewalling Freedom-of-Information-Act.” [↩]