Text chunks accidentally concatenated?
Dear contributor, first of all, thank you for offering such sample dataset from the originals. This speeds up the research for those just looking to grab a small to medium size text dump.
I have used this dataset for some annotation in a research group and unfortunately found out that many of the text chunks are a concatenation of two randomly selected texts, rendering the chunks very often completely incoherent. Sometimes, one of the chunks will have short length, which is not of a problem. However, very often, both chunks are of comparable size, which will impact many downstream tasks that rely on the coherence of the texts being used.
If this is mistake on my side, I apologize, however, I found some examples also in the data viewer here on HuggingFace. So, I suppose this is a problem in the dataset, unfortunely.
Here is an example: (id 261)
"""
Sadly, however, many of these initiatives have fallen victim to local authority funding cuts. The impact of austerity has been compounded by a loss of confidence – in which Putnam's research played its part – about fostering strong diverse communities. Multiculturalism has fallen from favour, misunderstood and maligned as the set of ideas that guided community relations for a generation. No one was more acutely aware of this danger than Putnam himself when he talked to me on the publication of his research in 2007, the timing made the danger all the more acute in the aftermath of 7/7 bombings. Since then the theme of integration has come to dominate – with its coercive and conformist overtones. The result has been ahad to fight Ahenobarbus like 4 times because he kept letting him go. With that, Caesar turned into an absolute master by making reasonable calculated risks and surviving them. By the point he was clearing the last of the Pompeans in Africa at the end of the Civil War, he didn't even leave his tent to give commands
"""