about summary refs log tree commit diff
path: root/corp/russian/data-import/src/mappings.rs
diff options
context:
space:
mode:
authorVincent Ambo <mail@tazj.in>2023-01-24T22·29+0300
committerclbot <clbot@tvl.fyi>2023-01-24T22·41+0000
commit80723b708d6edc44be00b61fb02260800101dcbb (patch)
tree81421796159849f3a9b82896a1dcd433783f3308 /corp/russian/data-import/src/mappings.rs
parent8d594658ab1b394a24f62528e761b54bacde482c (diff)
feat(corp/data-import): map OC lemma grammemes to OR form types r/5751
Change-Id: Ie804d185269336b0d9fe417754e5e795918e65b8
Reviewed-on: https://cl.tvl.fyi/c/depot/+/7923
Autosubmit: tazjin <tazjin@tvl.su>
Reviewed-by: tazjin <tazjin@tvl.su>
Tested-by: BuildkiteCI
Diffstat (limited to 'corp/russian/data-import/src/mappings.rs')
-rw-r--r--corp/russian/data-import/src/mappings.rs118
1 files changed, 103 insertions, 15 deletions
diff --git a/corp/russian/data-import/src/mappings.rs b/corp/russian/data-import/src/mappings.rs
index 5d8d4f3352fb..8a581ff86ba8 100644
--- a/corp/russian/data-import/src/mappings.rs
+++ b/corp/russian/data-import/src/mappings.rs
@@ -1,16 +1,103 @@
 //! Manual mapping of some data structures in OC/OR corpora.
 
-/// This maps the *names* of OpenRussian grammemes (the set of
-/// `form_type` fields in the `word_forms` table) to the *names*
-/// of OpenCorpora grammemes.
+/// Maps the *names* of OpenRussian grammemes (the `form_type` fields
+/// in the `or_word_forms` table) to the *set* of OpenCorpora
+/// grammemes attached to them corresponding lemma in the `oc_lemmas`
+/// table.
 ///
-/// The names of the OR grammemes are much easier to understand in
-/// general, as the OC ones seem to have strange acronyms in them,
-/// however the OC ones are much more structured.
+/// This *only* includes grammatical information about the lemma of
+/// the word (such as whether it is a verb or other type), but *not*
+/// information about the specific instance of the word (such as its
+/// gender).
 ///
-/// As these forms map to the word_forms table they lack the forms
-/// attached to the lemmata.
-pub const FORM_TYPES_GRAMMEMES: &'static [(&'static str, &'static [&'static str])] = &[
+/// Correctly corresponding these requires use of all mapping tables.
+pub const FORMS_LEMMATA_GRAMMEME_MAP: &'static [(&'static str, &'static [&'static str])] = &[
+    ("ru_adj_comparative", &["COMP"]),
+    ("ru_adj_superlative", &["ADJF", "Supr"]),
+    ("ru_adj_f_acc", &["ADJF"]),
+    ("ru_adj_f_dat", &["ADJF"]),
+    ("ru_adj_f_gen", &["ADJF"]),
+    ("ru_adj_f_inst", &["ADJF"]),
+    ("ru_adj_f_nom", &["ADJF"]),
+    ("ru_adj_f_prep", &["ADJF"]),
+    ("ru_adj_m_acc", &["ADJF"]),
+    ("ru_adj_m_dat", &["ADJF"]),
+    ("ru_adj_m_gen", &["ADJF"]),
+    ("ru_adj_m_inst", &["ADJF"]),
+    ("ru_adj_m_nom", &["ADJF"]),
+    ("ru_adj_m_prep", &["ADJF"]),
+    ("ru_adj_n_acc", &["ADJF"]),
+    ("ru_adj_n_dat", &["ADJF"]),
+    ("ru_adj_n_gen", &["ADJF"]),
+    ("ru_adj_n_inst", &["ADJF"]),
+    ("ru_adj_n_nom", &["ADJF"]),
+    ("ru_adj_n_prep", &["ADJF"]),
+    ("ru_adj_pl_acc", &["ADJF"]),
+    ("ru_adj_pl_dat", &["ADJF"]),
+    ("ru_adj_pl_gen", &["ADJF"]),
+    ("ru_adj_pl_inst", &["ADJF"]),
+    ("ru_adj_pl_nom", &["ADJF"]),
+    ("ru_adj_pl_prep", &["ADJF"]),
+    ("ru_adj_short_f", &["ADJS"]),
+    ("ru_adj_short_m", &["ADJS"]),
+    ("ru_adj_short_n", &["ADJS"]),
+    ("ru_adj_short_pl", &["ADJS"]),
+    ("ru_noun_pl_acc", &["NOUN"]),
+    ("ru_noun_pl_dat", &["NOUN"]),
+    ("ru_noun_pl_gen", &["NOUN"]),
+    ("ru_noun_pl_inst", &["NOUN"]),
+    ("ru_noun_pl_nom", &["NOUN"]),
+    ("ru_noun_pl_prep", &["NOUN"]),
+    ("ru_noun_sg_acc", &["NOUN"]),
+    ("ru_noun_sg_dat", &["NOUN"]),
+    ("ru_noun_sg_gen", &["NOUN"]),
+    ("ru_noun_sg_inst", &["NOUN"]),
+    ("ru_noun_sg_nom", &["NOUN"]),
+    ("ru_noun_sg_prep", &["NOUN"]),
+    ("ru_verb_gerund_past", &["GRND"]),
+    ("ru_verb_gerund_present", &["GRND"]),
+    ("ru_verb_imperative_pl", &["VERB"]),
+    ("ru_verb_imperative_sg", &["VERB"]),
+    ("ru_verb_past_f", &["VERB"]),
+    ("ru_verb_past_m", &["VERB"]),
+    ("ru_verb_past_n", &["VERB"]),
+    ("ru_verb_past_pl", &["VERB"]),
+    ("ru_verb_presfut_pl1", &["VERB"]),
+    ("ru_verb_presfut_pl2", &["VERB"]),
+    ("ru_verb_presfut_pl3", &["VERB"]),
+    ("ru_verb_presfut_sg1", &["VERB"]),
+    ("ru_verb_presfut_sg2", &["VERB"]),
+    ("ru_verb_presfut_sg3", &["VERB"]),
+    (
+        "ru_base",
+        &[ /* nothing consistent, except often 'Fixd' */ ],
+    ),
+    ("ru_verb_participle_active_past", &["PRTF", "past", "actv"]),
+    (
+        "ru_verb_participle_active_present",
+        &["PRTF", "pres", "actv"],
+    ),
+    (
+        "ru_verb_participle_passive_past",
+        &["PRTF", "past", "passv"],
+    ),
+    (
+        "ru_verb_participle_passive_present",
+        &["PRTF", "pres", "passv"],
+    ),
+];
+
+/// Maps the *names* of OpenRussian grammemes (the `form_type` fields
+/// in the `or_word_forms` table) to the *set* of OpenCorpora
+/// grammemes attached to them corresponding words in the `oc_words`
+/// table.
+///
+/// This includes grammatical information about the "instance" of the
+/// word (such as its gender), but *not* the higher-level type
+/// information about its lemma.
+///
+/// Correctly corresponding these requires use of all mapping tables.
+pub const FORMS_WORDS_GRAMMEME_MAP: &'static [(&'static str, &'static [&'static str])] = &[
     ("ru_adj_comparative", &["Cmp2"]),
     ("ru_adj_f_acc", &["femn", "sing", "accs"]),
     ("ru_adj_f_dat", &["femn", "sing", "datv"]),
@@ -59,12 +146,13 @@ pub const FORM_TYPES_GRAMMEMES: &'static [(&'static str, &'static [&'static str]
     ("ru_verb_past_m", &["masc", "sing", "past"]),
     ("ru_verb_past_n", &["neut", "sing", "past"]),
     ("ru_verb_past_pl", &["plur", "past"]),
-    ("ru_verb_presfut_pl1", &["plur", "1per", "pres"]),
-    ("ru_verb_presfut_pl2", &["plur", "2per", "pres"]),
-    ("ru_verb_presfut_pl3", &["plur", "3per", "pres"]),
-    ("ru_verb_presfut_sg1", &["sing", "1per", "pres"]),
-    ("ru_verb_presfut_sg2", &["sing", "2per", "pres"]),
-    ("ru_verb_presfut_sg3", &["sing", "3per", "pres"]),
+    // these also contain "pres" or "futr", depending on the verb.
+    ("ru_verb_presfut_pl1", &["plur", "1per"]),
+    ("ru_verb_presfut_pl2", &["plur", "2per"]),
+    ("ru_verb_presfut_pl3", &["plur", "3per"]),
+    ("ru_verb_presfut_sg1", &["sing", "1per"]),
+    ("ru_verb_presfut_sg2", &["sing", "2per"]),
+    ("ru_verb_presfut_sg3", &["sing", "3per"]),
     // Unclear items, probably only useful tags on lemmata
     (
         "ru_verb_gerund_present",