Difference between revisions of "Guarani/Universal Dependencies"
| Line 1: | Line 1: | ||
=Evaluation= | =Evaluation= | ||
| − | |||
| − | |||
| − | |||
| − | |||
==Corpora== | ==Corpora== | ||
{| class="wikitable" | {| class="wikitable" | ||
| Line 21: | Line 17: | ||
|} | |} | ||
| + | ==Withmorph== | ||
| + | Here are the results from using <code>gug.withmorph.udpipe</code> on the two conllu files: | ||
| + | |||
| + | {| class="wikitable" | ||
| + | |+Results | ||
| + | |- | ||
| + | | | ||
| + | |UAS | ||
| + | |LAS | ||
| + | |- | ||
| + | |<code>gug.annotated.ud.conllu</code> | ||
| + | | 91.14% | ||
| + | | 78.48% | ||
| + | |- | ||
| + | |<code>gug.annotated2.ud.conllu</code> | ||
| + | | 60.19% | ||
| + | | 47.57% | ||
| + | |} | ||
| + | |||
| + | ==Nomorph== | ||
| + | Here are the results from using <code>gug.nomorph.udpipe</code> on the two conllu files: | ||
| + | |||
| + | {| class="wikitable" | ||
| + | |+Results | ||
| + | |- | ||
| + | | | ||
| + | |UAS | ||
| + | |LAS | ||
| + | |- | ||
| + | |<code>gug.annotated.ud.conllu</code> | ||
| + | | 60.13% | ||
| + | | 50.00% | ||
| + | |- | ||
| + | |<code>gug.annotated2.ud.conllu</code> | ||
| + | | 33.01% | ||
| + | | 26.21% | ||
| + | |} | ||
| + | |||
| + | =Dependency Relations= | ||
| + | ==nsubj== | ||
| + | The nsubj relation is used to mark nominal subjects. In languages like English, we almost always have a word that acts as the nominal subject (occasionally there are implied subjects like in "Go!"). However, in Guarani, because person is marked on verbs as a prefix, there often sentences that do not have a distinct word that acts as a the nominal subject. | ||
| + | Consider the English sentence "I eat." In this sentence, we would mark "I" as the nsubj. However, in Guarani we could write the equivalent sentence as "Akaru" ("karu" = eat, "a-" = I), or as "Che akaru" ("che" = I). The first person pronoun is optional in this case. Because there is not a good way to mark a word as both a root and an nsubj, I ended up with some sentences that did not have an nsubj. Here are two example sentences: | ||
| + | * "Oguata." (He/She walks). In this sentence, "Oguata" is both the root and the nsubj. | ||
| + | * "Che aguata." (I walk). In this sentence, "che" is the nsubj and "aguata" is the root. | ||
| + | ==nummod== | ||
| + | The nummod relation came up fairly frequently in my corpora. I think this is mostly due to the fact that the word "peteĩ" (One) is often used as an indefinite determiner. Consider the following sentences: | ||
| + | *"Peteĩ jagua opo" (A/One dog jumps). In this sentence, it seems plausible to tag "peteĩ" as either a nummod or det. | ||
| + | *"Mokoĩ ava ajeroky" (Two men dance). In this sentence, "mokoĩ" (two) is a nummod for "ava" (man). | ||
| − | + | ==amod== | |
| + | The amod relation also came up fairly frequently in my corpora. The amod usually just behaves like it would in English. Adjectives are generally amods. Consider the following phrases: | ||
| + | *"ahy'o rasy" (sore throat). In this sentence, "ahy'o" throat is modified by the adjective "rasy" (sore"). Here, "rasy" is an amod. | ||
| + | *"Jakare kuña" (female crocodile). In this sentence, "jakare" (crocodile) is modified by the noun "kuña" (woman). Here, "kuña" is an amod, even though the lemma is a noun. | ||
| + | ==det== | ||
| + | Determiners were also frequent in the corpora. Guarani borrows the definite determiner "la" from Spanish. However, it is often optional. As mentioned early, the word "peteĩ" (one) often acts as an indefinite determiner. Consider the following sentences: | ||
| + | * "Maria okambia la istória" (Maria changes the story). Here, the word "la" (~the) acts as the definite determiner for "istória" (story). | ||
| + | * "Otopa peteĩ ita" (He/she throws a/one stone). Here, the word "peteĩ" (one) can be viewed as either an indefinite determiner for "ita' (stone), or as a nummod for "ita." | ||
| + | ==obj== | ||
| + | Direct objects in Guarani are usually marked with the suff "-pe." Sometimes, the suffix is not included (I couldn't find an explanation for when it is or is necessary). Consider the following sentences: | ||
| + | * "Peteĩ jagua ohecha mbarakajape" (A dog looks at a cat). Here, the word "mbarakaja" (cat) is behaving as the direct object of the verb "ohecha" (he/she/it looks), so "mbarakaja" gets the suffix "-pe." | ||
| + | * "Nde reguereko y" (You have water). Here, the word "y" (water) is the direct object of the root "reguereko" (you have). However, note that "y" does not have the suffix "-pe" in this case. | ||
[[Category:sp17_UD]] | [[Category:sp17_UD]] | ||
Latest revision as of 03:01, 26 April 2017
Contents
Evaluation
Corpora
| Number of sentences | Number of forms | |
gug.annotated.ud.conllu
|
33 | 158 |
gug.annotated2.ud.conllu
|
23 | 103 |
Withmorph
Here are the results from using gug.withmorph.udpipe on the two conllu files:
| UAS | LAS | |
gug.annotated.ud.conllu
|
91.14% | 78.48% |
gug.annotated2.ud.conllu
|
60.19% | 47.57% |
Nomorph
Here are the results from using gug.nomorph.udpipe on the two conllu files:
| UAS | LAS | |
gug.annotated.ud.conllu
|
60.13% | 50.00% |
gug.annotated2.ud.conllu
|
33.01% | 26.21% |
Dependency Relations
nsubj
The nsubj relation is used to mark nominal subjects. In languages like English, we almost always have a word that acts as the nominal subject (occasionally there are implied subjects like in "Go!"). However, in Guarani, because person is marked on verbs as a prefix, there often sentences that do not have a distinct word that acts as a the nominal subject.
Consider the English sentence "I eat." In this sentence, we would mark "I" as the nsubj. However, in Guarani we could write the equivalent sentence as "Akaru" ("karu" = eat, "a-" = I), or as "Che akaru" ("che" = I). The first person pronoun is optional in this case. Because there is not a good way to mark a word as both a root and an nsubj, I ended up with some sentences that did not have an nsubj. Here are two example sentences:
- "Oguata." (He/She walks). In this sentence, "Oguata" is both the root and the nsubj.
- "Che aguata." (I walk). In this sentence, "che" is the nsubj and "aguata" is the root.
nummod
The nummod relation came up fairly frequently in my corpora. I think this is mostly due to the fact that the word "peteĩ" (One) is often used as an indefinite determiner. Consider the following sentences:
- "Peteĩ jagua opo" (A/One dog jumps). In this sentence, it seems plausible to tag "peteĩ" as either a nummod or det.
- "Mokoĩ ava ajeroky" (Two men dance). In this sentence, "mokoĩ" (two) is a nummod for "ava" (man).
amod
The amod relation also came up fairly frequently in my corpora. The amod usually just behaves like it would in English. Adjectives are generally amods. Consider the following phrases:
- "ahy'o rasy" (sore throat). In this sentence, "ahy'o" throat is modified by the adjective "rasy" (sore"). Here, "rasy" is an amod.
- "Jakare kuña" (female crocodile). In this sentence, "jakare" (crocodile) is modified by the noun "kuña" (woman). Here, "kuña" is an amod, even though the lemma is a noun.
det
Determiners were also frequent in the corpora. Guarani borrows the definite determiner "la" from Spanish. However, it is often optional. As mentioned early, the word "peteĩ" (one) often acts as an indefinite determiner. Consider the following sentences:
- "Maria okambia la istória" (Maria changes the story). Here, the word "la" (~the) acts as the definite determiner for "istória" (story).
- "Otopa peteĩ ita" (He/she throws a/one stone). Here, the word "peteĩ" (one) can be viewed as either an indefinite determiner for "ita' (stone), or as a nummod for "ita."
obj
Direct objects in Guarani are usually marked with the suff "-pe." Sometimes, the suffix is not included (I couldn't find an explanation for when it is or is necessary). Consider the following sentences:
- "Peteĩ jagua ohecha mbarakajape" (A dog looks at a cat). Here, the word "mbarakaja" (cat) is behaving as the direct object of the verb "ohecha" (he/she/it looks), so "mbarakaja" gets the suffix "-pe."
- "Nde reguereko y" (You have water). Here, the word "y" (water) is the direct object of the root "reguereko" (you have). However, note that "y" does not have the suffix "-pe" in this case.