For obtaining the contexts in CoNLL format, we run context_process.py, as in the following example, for AJMC in English.
Before doing so, we need a file that contains all contexts in one, and thus:
cat datasets/ajmc/en/kb/en_wk5m_simple_time/HIPE-2022-v2.1-ajmc-train-en.kb datasets/ajmc/en/kb/en_wk5m_simple_time/HIPE-2022-v2.1-ajmc-dev-en.kb datasets/ajmc/en/kb/en_wk5m_simple_time/HIPE-2022-v2.1-ajmc-test_allmasked-en.kb > datasets/ajmc/en/kb/en_wk5m_simple_time/HIPE-2022-v2.1-ajmc-en.kb
And then:
python context_process.py \
--retrieval_file datasets/ajmc/en/kb/en_wk5m_simple_time/HIPE-2022-v2.1-ajmc-en.kb \
--conll_folder datasets/ajmc/en/ \
--lang en \
--version 3 \
--dataset ajmc
Which will generate the files in a folder hipe_doc_full_wiki_v3 in datasets/ajmc/en/.