The main criticism that I’ve heard regarding text analysis is something that we discussed in class, which is the idea that data such as word frequency doesn’t necessarily tell us anything larger about a work or primary source. I remember attending a talk where someone in the audience questioned the speaker’s methodology, as he asserted that word frequency didn’t necessarily mean that one work had significantly different themes or tones than another. I remember her answer being really good, essentially stating that the results weren’t the conclusion: the results simply helped guide her research.
Now that I’ve used Voyant, I understand much better what she meant. Plugging in the letters and memoirs of formerly enslaved people yielded fascinating results; I especially found the “Dreamscape” feature to be really interesting, as it charted routes across the globe to places mentioned in the texts. These results weren’t conclusions, but simply data that led to further research questions and directions. For example, the location Liberia was highlighted heavily in the Dreamscape feature, stating that it appeared in the texts 32 times, which is much higher than most other locations. This in of itself isn’t a conclusion; it makes sense that Liberia would be brought up a good deal in the 19th century in the context of the American Colonization Society, who sought to re-locate emancipated African Americans to Liberia. However, it is interesting that it appears so much more frequently than other places, and it might be really worthwhile to see the context in which it’s brought up across different sources. A really interesting research question could then develop from there.
As for cautions when using Voyant, I think it all comes back to using the data as a compass for research directions rather than simply conclusions. I think that what some people think when they hear the concept of text analysis is that tools like Voyant can tell a researcher how many times a certain word appears, and the researcher can make huge, sweeping generalizations from there. This is not the case, and I don’t think researchers who use text analysis are doing this. Rather, research questions arise from data, the same way that someone combing the archives for patterns and trends take their findings and expand on their research questions and interpretations rather than simply reporting what they found; as phrased in the article “Search and Replace,” text analysis, specifically distant reading, brings “criticism and data together.”
In conclusion, text analysis is not the raw production of data in order to make sweeping statements, but one more way in which to “mine” information for research and develop one’s thesis. It is part of a researcher’s interpretation the same way that quotes from a letter or census data are; in short, it serves as another form of methodology and source of evidence. One thing I’ve really been enjoying about digital humanities is how many different types of methodology it has been introducing me to, and this is just one more that could be useful in future projects.