Thursday, January 20, 2011

Openness and Culturomics

The Culturomics authors released a FAQ last week that responds to many of the questions floating around about their project. I should, by trade, be most interested in their responses to the lack of humanist involvement. I'll get to that in a bit. But instead, I find myself thinking more about what the requirements of openness are going to be for textual research.

That issue is deeply tied to some two-cultures questions about just what openness means. Matthew Jockers and Ted Underwood have been calling for a full release of the list of books behind the ngrams dataset. Culturomics (that's as clear as I can be on authorship, unfortunately) says they "have not received permission" yet to release the list of 5.2m books behind the set. I assume that's because Google's metadata is subject to proprietary restrictions from catalog aggregators and publishers—it will be interesting to see how they get out of that. Depending on what's in the metadata and its release format, that could range from barely readable to quite interesting.

On the other hand, as Culturomics points out, they have been commendably open with the ngrams data. They do far more than historians would have (I suspect) to make their experiments easily replicable, and their pages seem to indicate a plan to release more and more better cleaned data as time goes on. They seem to view the repository of data they're setting up as a field-changing contribution that will drive research in the quantitative study of culture. Files of the magnitude they're putting out are only possible for a very few organizations. Google is certainly the best positioned of those. If so, they're right to be so proud of their openness, and they're also right to have put that ahead of the bibliography. Cleaning and tokenizing textual data is a dreary task with enormous returns to scale, and if everyone needs basically the same dataset, research can move ahead faster with it even before the exact details are known.

But then again: if everyone needs the same dataset. That's a huge caveat, and it certainly isn't completely true. Some people want part-of-speech tagging on a representative corpus, and they'll want COHA or its descendants. Some people want very precisely edited texts of relatively canonical works, and they'll use MONK or WordHoard with the highly edited and tagged texts that come out of what Martin Mueller is calling digital lower criticism. Part of what happened when Culturomics came out to somewhat reserved enthusiasm is that the people interested in computer textual analysis—who already have systems in place and a clear idea of their needs—quickly realized that it didn't do what they needed, and that for many tasks (comparing versions of the folios? full part of speech tagging?), it might never make it.

At the same time, there are a lot of humanists who are still unclear on what, if anything, they can get out of lexical statistics. Some found ngrams eye-opening; some just found it fun; and some, I think, were put off a bit. First, by the scientific packaging; and second, by the lack of traditional humanist niceties like a bibliography, or some historiography, or a clear phrasing of what existing problems ngrams will solve. Since all humanists, by union contract, own the exact same 2008 white MacBook I have, most couldn't do anything much with the gigabytes of text files offered for download. You certainly can't open it with Excel to find what you want—even to get a basic wordcount for a span of years basically requires some sort of program. Casual humanists would probably be happier with less openness and more clarity—too much information can seem like a way of stonewalling, particularly when the information you most want isn't necessarily there. In any case, what they didn't necessarily get is a sense of the immediate applicability to live questions in the humanities.

That's because, in part, of the different types of openness. The openness of the sciences is based around replicability of experiments with relatively constrained goals; whereas currently, no one knows just what we're headed for in the humanities. (Particularly in history, which I promise a lot more about  later.) The openness of culturomics will let a thousand flowers bloom, but only in one type of research. Of course there will be other types: but the combined cultural capital of the route this project (Harvard-Google-Science-New York Times) makes it important to be clear that the release of centralized datasets, a la genomics, is not the long-term solution to allowing digital history.

I should be equally clear, though, that it is a long-term solution—some sort of baseline data is incredibly useful for all sorts of textual analysis, and whatever Google provides will probably be the best we get. I've had a little trouble so far figuring out how to use the Culturomics data to clean up my own dataset (largely because I don't want to adopt their ways of using capitalization and apostrophes for memory-saving reasons) but with a larger dataset, those sorts of problems should melt away. As they role out some more sets of genres with better metadata, it will provide an amazing group of genre baselines for comparison of more localized texts.

But given the limitations of ngrams (using the word generically) data, I tend to think that data's usefulness will rest not only in its openness a la genomics, but in its ability to complement other datasources. If Google ngrams is the best solution we can come up with for linking book metadata to textual data, we are going to be largely restricted to the studies of fame and repression in 'culture' that the culturomists have been releasing so far—studies that don't get to the core of most of the historiography which rely on a lot of different ways of thinking about how the web of language intersects at various levels.

Openness in the digital humanities needs to be about interoperability as well as replicability. Ngrams is stellar on the second, and merely good on the first. Moving forward, I wonder how we can do better.

I think that's all I really have to say for now aside from a couple reflections about copyright I'll post in a bit. But I should put up their response to the problem of lack of humanistic involvement in their project that I worried about, though I wasn't the first (nor was Menand, I'm sure):
2. Why were there no humanists involved in this project?

That's incorrect. Erez studied Philosophy at Princeton as an undergrad and did a master's degree in Jewish History working with Elisheva Carlebach. Two of our other authors, Joseph Pickett (PhD, English Language and Literature, UMichigan) and Dale Hoiberg (PhD Chinese Literature, UChicago) are the Executive Editor of the American Heritage Dictionary and the Editor-in-Chief of the Encylopaedia Britannica. In addition, we were in contact with many humanists throughout the life of the project.

But more than just wrong, it's irrelevant. What matters is the quality of the data and the analyses in the paper and what it means for how we think about a great variety of phenomena - not the degrees we happen to hold or not to hold. If what we seek is a serious conversation about this work, we shouldn't exclude anyone who has something significant and thoughtful to say. That would be a shame.
When I was researching a section of the Humanities Indicators about the "Humanities Workforce," one of the priorities was to be inclusive about the range of occupations—editors, secondary teachers, journalists, not to mention archivists, librarians, museum curators—who were professional humanists without a research university chair. There's no bright line, and that's good. I certainly don't want to write anyone out peremptorily.

No comments:

Post a Comment