Feat/readability metric

[Noel-Bastubbe]

No description provided.

[Tomic-Riedel]

Hey, Lisa tasked me with checking the PRs for framework conventions and I found a few things to address before we can merge this:

metis/metric/readability/__init__.py and README.md
Neither of these should exist inside a dimension folder. None of the other dimensions (completeness/, consistency/, etc.) have them, and it's important to keep that consistent. Please remove both files and move any relevant documentation to docs/ or the main README.

Class naming
I see that ReadabilityLLM and ReadabilityWordNet are the "real" classes, with readability_llm and readability_wordnet as thin alias subclasses at the bottom just for registry registration. The convention (which all four existing metrics follow) is that the metric class itself carries the {dimension}_{technique} name and there shouldn't be an intermediate PascalCase class at all. Could you flatten these into a single class each?

DQmetric values
DQmetric="llm" and DQmetric="readability_wordnet" don't follow the PascalCase short-name convention we use elsewhere (e.g. "NullRatio", "CountFDViolations", "DuplicateCount"). Please update to something like "LLM" and "WordNet".

Helper files in the dimension folder
llm_backend.py, scorers.py, and tokenization.py are utility modules sitting inside metis/metric/readability/. Was there a particular reason for keeping them there? If not, the convention is that helpers live in metis/utils/. In that case I'd suggest moving them there, possibly under a metis/utils/readability/ subfolder if they're readability-specific.

DQgranularity="schema"
This introduces a new granularity level that doesn't exist anywhere else in the framework right now. Before we add a new one I think we should loop in @lisehr to check whether a "schema" granularity level makes sense to add to the framework.

Docstrings
Both assess() methods are missing docstrings. The base class has a detailed template for exactly this, so please add them.

Thanks!

[lisehr]

Please shortly comment what this does and why you couldn't use the python tokenization function

[lisehr]

@Tomic-Riedel thanks for all the comments. We agreed that "schema" granularity make sense for readability since we might also have more schema metrics in the future. For now, we focused on the data level per default.

[lisehr]

Also here, please shortly comment what this class is about

[lisehr]

LGTM and free to merge together with the other files, the small comments can be fixed lateron