There has been some interesting banter over at Andrew Keen's Great Seduction blog around the fantasy of Google Apps developing a "virtual editor" (already christened "Google VE"). Since the time I spent working with Mark Stefik on editing the book reviews published in Artificial Intelligence accounted for some of my most memorable and enjoyable life experiences, I feel a need to elaborate a bit less frivolously from a practitioner's point of view. Ironically, when I started to reflect on the nature of my own practices, I discovered that I was falling back on the same framework that I had previously used in writing about the practices of musical performance, the medieval trivium of logic, grammar, and rhetoric. I cannot speak for how Mark approached his end of the operation (except for the fact that, because I was in Singapore, we could not have collaborated anywhere near as well as we did without really good electronic mail support); but, when I approached the manuscript of a review, particularly one that took the trouble to explore its topic in depth, I found that I had to "live" in all three of those trivium disciplines. Thus, if anyone is seriously fantasizing about a "virtual editor," it is important to account for each of those perspectives in terms of their capacity for accommodating technological support.
Grammar is probably the best place to begin. In the first place Microsoft has already demonstrated that it is at least feasible, although Word's grammar checking is so inferior to the spell checking that I ignore it almost entirely. One of the interesting things about natural language processing is the early discovery (going all the way back to Terry Winograd's doctoral work) that parsing a sentence is not necessary for "understanding" it (whatever the meaning of that word in scare quotes may be). Quite the contrary, because of issues like ambiguity, Winograd demonstrated that a syntactic analysis often has to draw upon hypotheses, if not results, arising from attempts at a semantic analysis. The Word grammar checker seems to be pretty ignorant of semantics, so it is no wonder that it is limited.
More important, however, is that, even if you establish that a sentence is "well formed" within the constraints of grammatical rules, as an editor, you may still want to recommend that the author change it. When I was a student, a favorite joke about language would be, "I can diagram it, but I can't understand it." I can still say that about a lot of the stuff I read (including, often, my own)! In other words not only are the rules of grammar not necessary for understanding the sentence; but also they are not sufficient! Editing within the discipline of grammar involves subtleties of judgment that go beyond the domain of syntactic analysis. Not only do I doubt that any "virtual editor" would be capable of making just judgment calls; but also I suspect that software analysis would not even be able to detect where those judgment calls would have to be made by the human editor who, presumably, would be using the virtual editor as a labor-saving tool.
Now, whatever may be bubbling beneath the surface at Google Labs, on the basis of what we see, Google lives in the world of logic; but, as just about anyone would point out, theirs is a rather impoverished world. I am sure that we all have our stories about why Boolean combinations of words and phrases do not make for particularly good semantic representations; and therein lies the rub. The emerging behavior of "Google rage" is a manifestation of the frustration that arises when what the user means just does not register with the capabilities of the search engine.
However, if Google were to argue that throwing a lot of smart researchers at the logic problem would eventually result in more useful software, my guess is that they are totally oblivious to the world of rhetoric, to the point that it would not even be on the radar in any brainstorming about a virtual editor. The good news is that I shall always have opportunities to make fun of the rhetorical ineptitude that emerges when Google faces the public. The bad news is that no one over there is likely to see any significance in my indulgences in this particular form of ridicule.
Nevertheless, there are issues of rhetoric that cannot be ignored, particularly when one is editing a particularly long manuscript, because, the longer the text gets, the less likely it is to be read in its entirety. This is where all three worlds converge. Taken together, the analysis of grammar, logic, and rhetoric help us prioritize what we read. It all goes back to the same things I had been writing about musical performance, the ways in which a reader can sort out the embellishing from the embellished. If the reader cannot quickly home in on those pieces of text that will be subjectively embellished, (s)he cannot make any sense out of the embellishment (and will probably then just bail on the whole text). It is that lack of the ability to prioritize that also can make search experiences so frustrating, both in how the query is expressed and in how the results are delivered.
Even in an age of Google and Wikipedia, how we read texts is still a good model for how we understand the world. Unfortunately, too much of our technological innovations choose to ignore this premise, perhaps because it is too difficult to be embraced by technology. Nevertheless, it is how we live in the world; and it cannot be ignored.
If you want to make fun of Google's linguistic analysis there is plenty of ammunition available in the shape of Google translate.
ReplyDeleteEl barco attravesta una cerradura How long does it take to fill a cerradura
Possibly the best Quires dormir con fosforo
Spanish in point of fact satifies Winograd's research. It is correctly parsed (on the whole) but does not understand any meaning. It does not know what type of "lock" or what type of "match" you are referring to.
i love your post. i love your observations. thanks for the insight!
ReplyDelete