3/24/2022 - A Quick Update on My Thoughts on Copyright
I wanted to revisit a question that I implied during the last post. We questioned whether data contains meaningful content. In short, of course
data contains meaningful content, otherwise there would be no point in gathering data! However, let's forget for a moment that our 'data'
contains strings of text, which convey the content of King's novels. Instead of releasing plain text copies of King's novels that we have
cleaned up, let's instead process the data further into some kind of structure that would be useful for data analysis. A dataframe in R could
be such a structure. Converting the plain text into such a structure is necessary for processing and does not yet remove any pieces from
the text, meaning all of the text is still preserved. This data structure could easily be exported as a raw csv file for others to utilize
for research. Would this csv file still be considered to be King's original text? Or would it be more apt to call it data, since no human
would realistically read an entire novel in such a format. One potential argument is that someone who knows how to handle strings could
splice apart this dataframe and reform the text back into a text format with relatively little effort. Therefore, despite the format, this
csv file is still more text than data and falls under copyright still.
However, for any kind of natural language processing (or really any kind of data analyses), there are multiple steps of pre-processing. For example, we need to remove stop words and calculate other low-level properties, such as word frequencies. Removing stop words immediately renders this text unreadable for humans and having word frequencies available is not information that any human would utilize in reading the given text. This structure feels a lot more like data now, and it is still raw enough that many higher-level analyses can still be built upon this information. We then decide to release this pre-processed csv file in a repository (ie one csv file for each novel). No one can realistically go back and turn these files into a complete text (this is actually a really interesting NLP problem itself; it would be tricky to accurately replace stop words for an entire text without any mistakes), which means that we are (theoretically) not at risk for violating copyright since we are now releasing data instead of written text. As long as our colleague trusts how we did this pre-processing, this would be a more effective way to share data for reproducibility.
This of course assumes that data derived from a fictional text is a separate entity than the text itself, which again, falls into a bit of a grey area: Does such data fall under copyright law? Does it fall under data privacy laws? Is it more similar to releasing a written textual analysis of the novel? Does it need to be its own category entirely?