Data loading/extraction pipeline

[borh]

Pinned issue on data ingestion/normalization pipeline.
As we want to keep the system as open as possible, we focus on open data as far as possible.

• Investigate using wtpsplit for sentence segmentation for datasets with ambiguous sentence/paragraph boundaries
• Investigate topic modeling extraction to be potentially used to show the topic breakdown of a collocation
• Integrate Huggingface datasets support. Initial implementation should load the Japanese Wikipedia from wikimedia/wikipedia, cf. https://huggingface.co/datasets/wikimedia/wikipedia?library=datasets --when implementing, use the streaming=True option of load_dataset to only load a sample for testing.
• Fix extraction bugs #8
• Expand extraction of current case relation triplets
• Expand extraction to 2-5 gram relations