Uh oh!
There was an error while loading. Please reload this page.
Implement BLEU score evaluation for NLP tests - #6537
Conversation
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Add a new library for algorithmic NPL scoring evaluators, named
Microsoft.Extensions.AI.Evaluation.NLP.The first such evaluator that is implemented is BLEU. For this implementation we default to 4 even weights for the n-gram comparisons and we use the smoothing method 4 from Chen and Cherry (2014) for sentence-level BLEU scores. These are the same defaults chosen by the Azure python evaluation SDK.
Also included is a simple word tokenizer based on the tokenizers used for other BLEU implementations, such as MOSES, SacreBLEU and NLTK.
Microsoft Reviewers: Open in CodeFlow