Add support for chunking_settings in semantic_text fields

[kderusso]

Updates the specification to include chunking_settings for semantic_text fields which were added in elastic/elasticsearch#121041

[pquentin]

Thank you for this! Before we review in more depth, I'm wondering if this should be consolidated with InferenceChunkingSettings or not.

[pquentin]

We have a similar class for the Inference API; should they be the same, or are they also distinct in Elasticsearch?

elasticsearch-specification/specification/inference/_types/Services.ts

Lines 60 to 89 in 1447a82

	/**
	* Chunking configuration object
	*/
	export class InferenceChunkingSettings {
	/**
	* The maximum size of a chunk in words.
	* This value cannot be higher than `300` or lower than `20` (for `sentence` strategy) or `10` (for `word` strategy).
	* @server_default 250
	*/
	max_chunk_size?: integer
	/**
	* The number of overlapping words for chunks.
	* It is applicable only to a `word` chunking strategy.
	* This value cannot be higher than half the `max_chunk_size` value.
	* @server_default 100
	*/
	overlap?: integer
	/**
	* The number of overlapping sentences for chunks.
	* It is applicable only for a `sentence` chunking strategy.
	* It can be either `1` or `0`.
	* @server_default 1
	*/
	sentence_overlap?: integer
	/**
	* The chunking strategy: `sentence` or `word`.
	* @server_default sentence
	*/
	strategy?: string
	}

[kderusso]

Oh nice catch!

The only concern that I have with combining them, is that we currently require all required settings to be sent in, with no defaults applied. So if you select word based chunking and don't send in both max_chunk_size and overlap you will get an error. This made sense when we wrote it because you're already putting in custom overrides, though I suppose we could revisit this in the future.

Based on this specification, though, max_chunk_size and overlap would both be optional due to default values, and using this would break.

WDYT the best course of action is? Maybe put in a comment why they're separate and keep them separate for now?

[pquentin]

Right, let's keep them separate. To ensure they keep using the same property names, we can use OverloadOf: https://github.com/elastic/elasticsearch-specification/blob/main/docs/behaviors.md#overloadof. With that, we don't need a comment!

[kderusso]

TIL, thanks!

[pquentin]

Thanks for making the changes. However, were the parameters different? You used to have type and it's now strategy.

[kderusso]

That would have been a bug 🤦‍♀️ Here's the PR with the example, strategy is correct: elastic/elasticsearch#121041

[pquentin]

Thanks! LGTM.