Analysis of Speeches Using Machine Learning

Recently, I discovered a very useful tool when I am learning computational anthropology. It is called Natural Language Processing, or NLP, using a machine learning software named Watson, created by IBM. What it does is that it can identify the emotions and relevance of the words in a given paragraph. Hence, I used it to analyse and compare Martin Luther King's and Donald Trump's speeches.

Donald Trump said, "There’s no greater crisis facing our nation today than the catastrophe on our Southernborder. You’ve been watching it and a lot of the fake news. And you got a lot of them back …"

I think IBM Watson NLU is a really powerful tool to analyse these speeches. It had very similar understandings, relevant concepts and ratings as I did when I finished reading these 2 speeches, but it provides a clearer and much more logical analysis about these 2 speeches. Such as relevance of many specific concepts.

For Trump's speech, it has got similar sentiments scores as I did when I finished the reading (51% sadness, 17% joy, 48% anger, and 11% fear). However, the ML model does have limitations on the accuracy of interpretation of human emotions. President Trump did criticize about the White House, but it is clear he want it to be a better place, and this is a positive emotion. IBM Watson has trouble finding that emotion because it only sees through the literal meanings of the speeches or the paragraphs.

What's more, in my opinion, IBM Watson has trouble finding the relationship among different concepts when it is evaluating emotions. For instance, Donald Trump dislikes the idea of president Joe, but Watson analyses it as he dislikes the concept president, and feels neutral about the word "Joe". This is clearly a misinterpretation of Trump's feelings. They should be linked together, as "president Joe".

For Martin Luther King's speech, IBM Watson did a good job finding the relevance of different concepts.

However, I do realize some difference in the understanding of emotions of certain concepts between Watson and myself. The phrase "bright day of justice" should be positive and filled with hope, but Watson gave a negative rating on this concept. What's more.

There should be less percentage of joy (Watson gave a 65% joy rating) in his speech. In my opinion, he was more upset and angry instead of being happy to present such inequality in front of the nation.

I actually tried to use Subdown web and IBM Watson NLP trying to analyse some of the YouTube news channels and BBC news. I actually had some interesting findings. If you are reading my analysis, you can also try to analyse the political biases YouTubers have using this powerful tool.