In evaluating the performance of various natural language preprocessing systems, particularly for Polish, non-neural tools like Concraft and UDPipe performed comparably to neural network systems in specific tasks, particularly segmentation.
Tagset selection shows significant impact on system performance; systems trained on the Morfeusz tagset generally underperformed in comparison to those trained on the UD tagset, highlighting sensitivities in tagset configurations.
Collection
[
|
...
]