Machine Learning and the Humanities

17, సెప్టెంబర్ 2014, బుధవారం

Machine Learning and the Humanities

Red Vineyards, Vincent van Gogh (source)

This past summer, I gave a paper at the Grief. Language. Art. Conference as part of the Embodiments Research Group of the University of Liverpool, where I discussed natural language processing and some machine learning algorithms to accomplish sentiment analysis. I utilized Stanford’s Java-based NLP tools to demonstrate where we are in terms of sentiment analysis and what we may be able to achieve soon.

My paper, “Parsing Grief through Sentiment Analysis,” was concerned with how sentiment analysis is largely built on social media marketing tools, and does not move much beyond positive / negative polarity at the moment (can a computer parse grief?), though the poetry I was demonstrating, care of Sir Thomas Wyatt, did manage to be assessed as “Very Negative” at times, to the delight of my fellow conference delegates.

Screen shot of NLP on the command line.

A recent paper, concerning both computer science and art history, shows that there is a lot of potential for machine learning to assess art pieces. Babak Saleh, Kanako Abe, Ravneet Singh Arora, and Ahmed Elgammal, the authors of “Toward Automated Discovery of Artistic Influence,” worked in computer vision and pattern recognition to determine whether computers can measure influence between works of art. The results, summed up here a bit hyperbolically, are pretty interesting as the algorithm seems quite capable of recognizing pattern and influence to an extent.

Among the links it made were between Frederic Bazille’s Studio 9 Rue de la Condamine (1870) and Norman Rockwell’s Shuffleton’s Barber Shop (1950), which the authors claim has not been written about previously. There is certainly a compositional similarity with shared objects and architectural elements:

Posted on the site in accordance with fair use principles.

I would love to know what Art Historians think about these claims and whether they agree with the assessments. Ideally, there would be more collaboration between computer scientists and humanities scholars to work on these tools that concern the latter’s research.

Certainly, this kind of work can be a gateway to new methodologies to be employed by humanities scholars, and I hope that my colleagues do not fear a singularity will occur, as machine learning is not true learning but only pattern recognition. I think at this stage it can be problematic to be sure, but computer algorithms can do a lot of work over large corpora that researchers will not be able to make a dent in during their lifetimes. It is in this way that we would be able to see the general sentiment of Shakespeare’s texts in comparison to his peers, or look at general influences within an entire century’s work of paintings.

Unknown 12:16 PM Komentar