We’re busy preparing to launch LingPipe 4.0. We’re leaning toward switching from our “unauthorized” Alias-i Royalty Free License (version 1) to the GNU Affero General Public License (version 3) (AGPLv3).
The motivation for our license was very similar to Affero’s. Affero, the original license developer, wanted to close the so-called “Application Service Provider (ASP) loophole”, which didn’t count running software as a service as redistribution. Thus application service providers (ASPs) may run Gnu Public Licensed (GPLed) software as a service without distributing their software own software linked to it. The AGPL is stronger (i.e. its viral license is more contagious) in that it considers service providers to be distributing the software.
We’re neither lawyers nor software freedom fighters (though I am getting tired of all this software patent nonsense). The motivation for our royalty-free license was partly to prevent potential paying customers from running LingPipe on proprietary data, while keeping the system as “open” as possible, especially for research and teaching purposes. Our business model was thus something like MySQL’s business model. Here’s a nice survey of open-source business models — we’re a mix of 3, 4, and 5, plus 1 if you count DARPA/NIH grants as donations/subsidies.
Check out Bruce Perens‘s quite sensible review of the plethora of OS licenses: How Many Open Source Licenses Do You Need? (Spoiler: Perens, who wrote the original version of OSI’s Open Source Definition now thinks 4 are sufficient; I suppose you can’t blame a computer scientist for generalizing!).
We’re wondering what the ramifications of a switch to AGPLv3 would be. On the upside, we will potentially attract more users and perhaps even developers by having a license that plays nicely with others (specifically GPL and Apache/BSD). We also think it may be more attractive to funding agencies, like NIH. On the downside, we’re worried it may adversely affect our ability to sell our software, or perhaps even the business as a whole.
Are we crazy, or should we have done this ages ago?
March 4, 2009 at 5:49 pm |
You’re not crazy – I think this sounds like a good idea. While I’m typing here, though, I should mention that I’ve never used LingPipe in a class because I feel like the learning curve is pretty steep… Any chance you’ll offer a tutorial at an upcoming (North American) ACL about how to incorporate LingPipe into a class?
March 4, 2009 at 6:00 pm |
The change makes sense to me. Though I have to wonder how many folks would deliberately try to subvert the intent of your non-commercial license, knowing that doing so would earn your enmity and surely your unwillingness to offer any kind of support.
March 4, 2009 at 6:24 pm |
@Rich W: What kind of class are we talking about?
There are at least four learning curves: comp ling, stats, Java, and LingPipe. The problem is that LingPipe presupposes both a reasonable understanding of stats and computational linguistics and a fairly strong grasp of Java idioms such as serialization and generics.
For an intro class for people who don’t already know Java pretty well, Bird et al.’s forthcoming book on NLTK in Python is probably more appropriate. For those also trying to get a start with search, Konchady’s book on Lucene, LingPipe and GATE is a nice overview, but also presupposes users know Java.
We’d love to do a LingPipe tutorial, but I’m afraid ACL tutorials tend to be on more technical topics than how to use pieces of software. Maybe I could submit something to the teaching workshop, but I haven’t taught comp ling in over ten years!
March 5, 2009 at 5:26 pm |
I guess I was thinking undergrad NLP or IR courses… I say that because you wrote “especially for research and teaching purposes” and though it’s occurred to me to try using LP in a class, I’ve never thought I could do it… so when you mentioned teaching purposes, I thought maybe you had some ideas about how to do that.