On Mon, Jan 26, 2009 at 3:47 AM, Petr Kadlec petr.kadlec@gmail.com wrote:

...

Is the code available and I have missed it? Do we have any other implementation?

I tried to do something similar (two examples are at http://mormegil.info/wp/blame/AIM.htm http://mormegil.info/wp/blame/AFC_Ajax.htm); the code is nothing secret, even though it is not too clean, and there is also no rocket science [you have been warned: https://opensvn.csie.org/traccgi/MWTools/browser/MWTools/trunk/PageBlame]:

I also have a blame engine of my own design. It is new and I haven't released the source.

The biggest problem I see with such tools is that it is IMHO unusable for any copyright-related purposes. My tool works by diffing the article revisions and tracking who was the last author of every word. Even though you can be much smarter than that, I don't believe you would be able to track all copyright-relevant contributions with that. As an example, consider using that tool on an article that was created by:

1. Importing an article with all its history from the English

Wikipedia to some other-language wiki. 2. Translating it into the local language (for more fun, imagine a language using a different script, e.g. Russian, or even Chinese)

There is IMHO no way the blame tool could track copyright properly through the translation (which it has to, copyright-wise). And even in the general case, I believe such tracking would be an AI-hard task (often, even a human is unable to do it properly…). Of course, such Blame tools are great for many reasons (which is why I wrote them), but I think the current context (license change, attribution etc.) does not fit them at all.

I think I have a more positive view than you do. Blame engines as a tool can certainly inform copyright discussions and provide relevant information, even though I agree they aren't by themselves a complete solution.

For example, with situations where one is trying to list a fixed number of "major authors" (as provided in the GFDL, for example), blaming tools can make a reasonable guess at which authors are relevant. They also help estimate the answer to important meta questions, such as "How many authors does a typical Wikipedia article really have?"

When the license calls for attribution to be treated in a "reasonable" way, I suspect that one could make a good case that relying on a good blame engine would often generate a reasonable attempt at attribution, even though there are cases (like translation) where they will fail. Attribution generated by blaming can be a good starting point, though it may not necessarily be the final answer.

-Robert Rohde

It's more of a starting point, to flag editors who may have made the edits. All that would remain is looking over if that user did in fact make that edit (and if they didn't, it's back to square 1)

On Jan 26, 2009, at 3:38 PM [Jan 26, 2009 ], Gregory Maxwell wrote:

On Mon, Jan 26, 2009 at 1:41 PM, Robert Rohde rarohde@gmail.com wrote: [snip]

...

When the license calls for attribution to be treated in a "reasonable" way, I suspect that one could make a good case that relying on a good blame engine would often generate a reasonable attempt at attribution,

[snip]

Often, sure— But what happens when it fails and you have someone yelling loudly on the talk page "Hey! it's misattributing my authorship to some dumb bot, yet I wrote the whole thing!" ...

It's not reasonable by any human (or legal) standard to continue to misattribute in a case like that, yet addressing that case with some automatically generated report is not easy.

(and, of course, it's a great starting point… so long as someone remembers to continually point out that its not a final answer).

---

Wikitech-l mailing list Wikitech-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Platonides wrote:

Gregory Maxwell wrote:

...

It's not reasonable by any human (or legal) standard to continue to misattribute in a case like that, yet addressing that case with some automatically generated report is not easy.

There could be an override for some articles for outsiders, just to stop misattribution in the time being, but it can't be treated as a solution. If it attricutes a bot, the algorith is wrong and shall be fixed.

This particular case should be easy to fix, but only because we're fairly meticulous about flagging bot accounts. Essentially, we'd be falling back to a human saying "that account is a bot, don't attribute anything to it".

Which, I suppose, is a fairly straightforward and objective decision to make, but it's still a human decision made according to someone's personal point of view. Certainly such a fix won't easily generalize to more controversial cases.

Add it to the tests and improve the algorithm. One very important point of having an automated way of dealing with this is precisely that it allows us to be completely neutral. The "author reviewer" doesn't have a POV, it doesn't care that the contributor is named "Willy on wheels" or "Gmaxwell".

Sort of reminds me about the story somewhere in the Jargon File about closing one's eyes so that the room would be empty.

Deciding who are the "major contributors" to an article is ultimately a subjective issue; it can be algorithmically approximated, given a suitable weighting function (such as "number of words contributed"), but ultimately the choice of the weighting function itself is to an extent a matter of personal opinion. Trying to achieve neutrality by focusing solely on purely objective metrics is likely to produce results that are neutral only in the sense that everyone can agree that they suck.

The hard part is obviously how to get that magic attribution tool. If at least there would have been some convention on summary syntax to be used when attributing someone else...

Such a convention (or, better yet, a separate field for attributing edits to someone else) would help a lot. Of course, it would only be as reliable as the users entering the data, and wouldn't help with legacy edits.

A particularly common and tricky use case would be text copied from one article to another (possibly across different wikis). Either the blame tool would have to chase such references (which seems impractical at least in cross-wiki cases) or it would have to rely on the users making the copy to correctly determine the actual author(s) of the copied text (a process which is highly error prone, even with computer assistance, and likely to lead to errors accumulating as text misattributed once is further copied using the incorrect attribution).

Also, there would obviously have to be some way to correct attribution errors after the fact. Which, of course, leads to the question of who should be authorized to make such corrections, and how this would be any less subjective than revising the author list directly.