Alignment

evaluatio.alignment¶

Functions¶

optimal_alignment¶

optimal_alignment(references: Iterable[object], hypotheses: Iterable[object]) -> List[_bindings.Alignment]

Compute an optimal alignment between a reference and hypothesis sequence.

Aligns each element in references to a span in hypotheses using dynamic programming, minimising the total edit distance between the two sequences. The result describes, for each reference index, which range of hypothesis indices it aligns to.

Parameters

• references : Iterable[object]
  Reference token sequence.

• hypotheses : Iterable[object]
  Hypothesis token sequence.

Returns

• List[Alignment]
  One Alignment per reference token, sorted by reference index. Each Alignment has three fields:

  • index : int — index of the token in references.

  • start : int — start index of the aligned span in hypotheses (inclusive).

  • end : int — end index of the aligned span in hypotheses (exclusive).

Note

• Tokens are compared using the same underlying mechanism as universal_edit_distance.

• Multiple optimal alignments might exist, one is returned arbitrarily, but deterministically.