Enum BleuSmoothing 

pub enum BleuSmoothing {
    None,
    AddEpsilon(f64),
    Exponential,
}

Smoothing method for BLEU score computation.

Sentence-level BLEU often produces zero scores when higher-order n-grams have no matches. Smoothing techniques address this by assigning small non-zero values to zero-count precisions.

References

Chen & Cherry, “A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU”, WMT 2014.

Variants

None

No smoothing. Zero precision at any n-gram order produces a zero overall score (standard corpus-level BLEU behaviour).

AddEpsilon(f64)

Add a small constant (epsilon, default 0.1) to zero-count precisions. Corresponds to method 1 in Chen & Cherry (2014).

Exponential

Exponential decay: for each n-gram order with zero matches, double a running multiplier k (starting at 1 and doubling on every zero) and replace the precision with 1 / (k * total_n). Corresponds to method 3 in Chen & Cherry (2014) and the default smoothing in SacreBLEU for sentence-level BLEU.

Trait Implementations

impl Clone for BleuSmoothing

fn clone(&self) -> BleuSmoothing

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for BleuSmoothing

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Default for BleuSmoothing

fn default() -> BleuSmoothing

Returns the “default value” for a type.