improved alignment "caching"

[keighrim]

New Feature Summary

As a app developer, I would like to have the superpower to navigate Alignment annotations and their arguments (target, source) more easily, instead of writing complicated graph building code every time I have to deal with alignments like

• input MMIF spec for RFB app app-role-filler-binder#2.
• text "slicer" based on time interval #280
  and many more instances.

implementation proposal

Add a hidden (ignored-when-serialized) attribute to each mmif.serialize.annotation.Annotation instance to hold a dict from Alignment object long_id to the counter part object long_id. For example, supposed that we have three annotation objects in a MMIF

{
  "metadata": {
    "mmif": "http://mmif.clams.ai/1.0.0"
  },
  "views": [
    {
      "id": "v_0",
      "metadata": {
        "timestamp": "2023-07-25T19:55:56.693173",
        "app": "http://apps.clams.ai/whisper-wrapper/v3",
        "contains": {
          "http://vocab.lappsgrid.org/Token": {},
          "http://mmif.clams.ai/vocabulary/TimeFrame/v1": {}, 
          "http://mmif.clams.ai/vocabulary/Alignment/v1": {},
          ...
        }
      },
      "annotations": [
        {
          "@type": "http://mmif.clams.ai/vocabulary/TextDocument/v1",
          ...
        },
        ...
        {
          "@type": "http://vocab.lappsgrid.org/Token",
          "properties": {
            "word": " Good",
            "start": 0,
            "end": 5,
            "document": "v_0:td_1",
            "id": "to_1"
          }
        },
        {
          "@type": "http://mmif.clams.ai/vocabulary/TimeFrame/v1",
          "properties": {
            "frameType": "speech",
            "start": 98.52,
            "end": 99.12,
            "id": "tf_1"
          }
        },
        {
          "@type": "http://mmif.clams.ai/vocabulary/Alignment/v1",
          "properties": {
            "source": "tf_1",
            "target": "to_1",
            "id": "al_2"
          }
        }, 
        ...
      ]
    }
  ]
}

When de-serialized, the annotation object will have hidden attributes to "cache" aligned annotation ids.

>>> m = Mmif(mmif_str)
>>> m.get('v_0:to_1')._alignments
{"v_0:al_2": "v_0:tf_1"}
>>> m.get('v_0:tf_1')._alignments
{"v_0:al_2": "v_0:to_1"}
>>> m.get('v_0:al_1')._alignments
{}  # unless alignment annotation itself is aligned to other annotations

If there are multiple alignments to an annotation, the dict simply needs to be extended with more. Updating the _alignments attribute (tentative name) should be added to

1. MMIF de-serialization (

   mmif-python/mmif/serialize/mmif.py

   Line 189 in 90ec32f

   def _deserialize(self, input_dict: dict) -> None:

   )
2. View::add_annotation method (

   mmif-python/mmif/serialize/view.py

   Line 106 in 2a5bffc

   def add_annotation(self, annotation: 'Annotation', overwrite=False) -> 'Annotation':

   )
3. View::new_annotation method (

   mmif-python/mmif/serialize/view.py

   Line 77 in 2a5bffc

   def new_annotation(self, at_type: Union[str, ThingTypesBase], aid: Optional[str] = None,

   )

limitation

Currently View object doesn't have access to its "parent" Mmif object, so getting long_ids of annotation object might not be trivial, unless we add that pointer to the parent in the View class.

Related

No response

Alternatives

No response

Additional context

No response