Relaxed criteria

[fginter]

Allow a basic CoNLL-U format check without the extra character set and symbol list restrictions imposed by UD.

[martinpopel]

As discussed elsewhere, this format only check could either

• be the default if no --lang is provided. With the online validation (where the proper --lang is used), I think there is no risk of releasing an invalid treebank. Or
• introducing a new option --formatonly (or something like that)

It is worth considering which tests should be included in the format-only version. We agree lang-specific deprels and spaces in forms/lemmas should be allowed. What about

• the empty-node-without-dependent test (see #11) or
• SpaceAfter=No matching the raw text?

[dan-zeman]

The new validator can test on 5 levels (the --level option):

1. Only backbone CoNLL-U. Fields can be empty, except for the ID column.
2. Universal-level. UPOS, HEAD, DEPREL and DEPS must have expected values. Language-specific extensions are not checked and all are allowed (i.e., there can be any feature-value pair in FEATS).
3. Universal content-based tests. E.g. conj must go left-to-right.
4. Language-specific features, deprels, tokens with spaces. You can still use lang "ud" with this level to require that there are no language-specific extensions.
5. Language-specific content-based tests. E.g. list of lemmas that can be auxiliary verbs in the language.