In the process of coding all of the Levant records, I found two additional places, both associated with code using get_loccodes(), where the program crashed due to finding a tuple where it expects a string (the program actually bombs at the point of a print statement so it confusingly manifests itself as Unicode error). This is likely either due to handling the compounds, or another variant on truly weird CoreNLP parsing. Rare -- 60 occurrences in a file of 130,000 sentences -- but reproducible and probably should get tracked down. These will be in the commits I did at the end of May-15.
In the process of coding all of the Levant records, I found two additional places, both associated with code using get_loccodes(), where the program crashed due to finding a tuple where it expects a string (the program actually bombs at the point of a print statement so it confusingly manifests itself as Unicode error). This is likely either due to handling the compounds, or another variant on truly weird CoreNLP parsing. Rare -- 60 occurrences in a file of 130,000 sentences -- but reproducible and probably should get tracked down. These will be in the commits I did at the end of May-15.