The most abundantly clear message to come out of the readings for this week is that text mining cannot and should not replace traditional historical methods, but that it should be used as a means of complimenting those traditional methods. It provides a means of dealing with what is becoming an increasingly unwieldy amount of historical information. This is made clear by the Miki Kaufman piece on quantifying Kissinger that uses text mining to poor through the “memcons” and “telcons” from Kissinger’s 17,500 meetings. that is a fantastic amount of information to try to make sense of. The problem is not limited to the 20th century.
Text mining can provide evidence about a mass of texts that we then look at more closely. In other words, through distant reading, we find areas to read more closely. Where text mining seems to me to have the greatest impact, however, is through generating new questions by finding trends one may not otherwise find. A good example of this is in Robert Nelson’s “Mining the Dispatch: Introduction,” in which the graphs generated from his topic modeling show an increase in runaway slave ads in 1562 causing him to ask if an increase in chaos from war mobilization in and around Richmond provided slaves with increased opportunity to runaway in larger numbers. From here, he would have to bring traditional methods to bear on the question. The point, however, is that text mining and topic modeling made it clear that there was a question that needed answering.
Text mining can provide another analytic approach that historians bring to bear on their available sources. However, we must remember that it does not provide conclusive proof, but another type of evidence to demonstrate a point or a new way to formulate questions.