Merge branch 'main' of github.com:disi-unibo-nlp/nlg-metricverse

nlgmetricverse/metrics/chrf/README.md

@@ -8,8 +8,8 @@ ChrF compares character n-grams in the reference and candidate sentences, instea
 The precision and recall are computed over the character n-grams for various values of n (upto 6) and are combined using arithmetic averaging to get the overall precision (chrP) and recall (chrR) respectively.
 In other words, chrP represents the percentage of matched character n-grams present in the hypothesis and chrR represents the percentage of character n-grams in the reference which are also present in the hypothesis.
 The final chrF score is then computed as:<br>
-<img src="https://render.githubusercontent.com/render/math?math={chrF_{\beta} = (1 + \beta^2) \frac{chrP chrR}{\beta^2 chrP + chrR}}##gh-light-mode-only">,<br>
-where <img src="https://render.githubusercontent.com/render/math?math={\beta}##gh-light-mode-only"> indicates that recall is given <img src="https://render.githubusercontent.com/render/math?math={\beta}##gh-light-mode-only"> times more weightage than precision.
+$$chrF_{\beta} = (1 + \beta^2) \frac{chrP chrR}{\beta^2 chrP + chrR}$$
+where $\beta$ indicates that recall is given $\beta$ times more weightage than precision.
 Popovic propose two enhanced versions of chrF: (i) chrF+, which also considers word unigrams; (ii) chrF++, which considers word unigrams and bigrams in addition to character n-grams.
 
 We use the implementation that is already present in sacreBLEU.
@@ -106,4 +106,4 @@ print(scores)
 ```
 
 ## Further References
-- See the [sacreBLEU README.md](https://github.com/mjpost/sacreBLEU#chrf--chrf) file for more information.
+- See the [sacreBLEU README.md](https://github.com/mjpost/sacreBLEU#chrf--chrf) file for more information.