Skip to content

Latest commit

 

History

History
38 lines (24 loc) · 1.03 KB

README.md

File metadata and controls

38 lines (24 loc) · 1.03 KB

CountBigramFreqInConlluCorpus

Count Bigram frequency in a conllu format corpus

1)Place your conllu files into directory named Texts

2)Launch count_bigram_freq.py script

2.1)Script works for large files (reads them in 1gb chunks)

2.2)Script creates 2 json files:

bigram_freq.json - dictionary of bigram frequency in corpus
unigram_freq.json - dictionary of unigram frequency in corpus

3)Optional:

sort_freq_dict.py - sorts dicts by frequency and creates freq_sorted.json files

CalculateTscoreForBigramsBasedOnFrequency

Calculate Tscores for bigrams based on their frequency

1)Launch bigrams_t_score.py script

1.1)If you want to use sorted dicts replace filenames in script to sorted

1.2)Script creates 1 json file:

bigram_t_scores.json - dictionary of bigram t_scores

Formula used:

$$\frac{B_f - \frac{U1_f + U2_f}{N}}{\sqrt{B_f}}$$
  • $N$ - words in corpus (sum of unigram values/frequencies in our case)
  • $B_f$ - bigram frequency
  • $U1_f$ - 1st unigram frequency
  • $U2_f$ - 2nd unigram frequency