Difference between revisions of "Lingala/Universal Dependencies"

From LING073
Jump to: navigation, search
(→‎obj)
 
(12 intermediate revisions by the same user not shown)
Line 1: Line 1:
−
Lingala Universal Dependencies are documented on this page.
+
Lingala Universal Dependencies are documented on this page. [https://github.swarthmore.edu/apowell1/ling073-lin-corpus Related models and corpora] are available on GitHub.
  
 
==Evaluation==
 
==Evaluation==
Line 12: Line 12:
 
|-
 
|-
 
! lin.annotated.ud.conllu
 
! lin.annotated.ud.conllu
−
|  
+
| UAS: 69.85%, LAS: 61.45%
−
|
+
| UAS: 61.83%, LAS: 53.44%
−
|
+
| 262
−
|
+
| 12
 
|-
 
|-
−
! lin.annotated.nomorph.conllu
+
! lin.annotated2.ud.conllu
−
|  
+
| UAS: 39.82%, LAS: 21.24%
−
|
+
| UAS: 29.20%, LAS: 20.35%
−
|
+
| 113
−
|
+
| 8
 
|}
 
|}
  
 +
The lower stats for <code>lin.annotated2.ud.conllu</code> are to be expected since the text for this corpus was drawn from a completely different source with a very different style.
  
 
==Dependency Relations==
 
==Dependency Relations==
 +
Examples of dependency relations are shown below. Note that Lingala has some odd constructions with nouns and verbs that can allow them to function as if they were a different part of speech, so sometimes a certain dependency might seem to correspond to an otherwise incompatible part of speech.
 +
 +
===ccomp===
 +
The ccomp relation is used in the manner described in the UD documentation: it denotes a clause relating to a main verb that is a core argument. This comes up very frequently in the Book of Mormon corpus that was being used for annotations, since the construction "It came to pass that..." is used quite frequently. The verb following "that" has a ccomp dependency to "came."
 +
 +
For a basic example, consider the following partial sentence:
 +
<pre>
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<esalemaki>"
 +
"kosalema" v urp p3 sp nn @root #2->0
 +
"<ete>"
 +
"ete" cnjsub @mark #3->4
 +
"<azongaki>"
 +
"kozónga" v urp p3 sg aa @ccomp #4->2
 +
</pre>
 +
 +
This is translated "It came to pass that he returned..." Here, "returned" has a ccomp dependency to "came." For a slightly more complicated example, consider the following excerpt:
 +
 +
<pre>
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<esalemaki>"
 +
"kosalema" v urp p3 sp nn @root #2->0
 +
"<ete>"
 +
"ete" cnjsub @mark #3->8
 +
"<bo>"
 +
"bó" cnjsub @mark #4->5
 +
"<etangaki>"
 +
"kotánga" v urp p3 sp nn @advcl #5->8
 +
"<ye>"
 +
"yé" prn pers p3 sg aa @nsubj #6->5
 +
"<,>"
 +
"," cm @punct #7->5
 +
"<atondisamaki>"
 +
"kotónda" v caus pass urp p3 sg aa @ccomp #8->2
 +
</pre>
 +
This is translated "And it came to pass that as he read he was filled..." As before, "filled" has a ccomp dependency to "came," but in this example there is an intervening advcl.
 +
 +
===nmod===
 +
The nmod relation is used quite frequently. Most frequently, it indicates possession, since possession in Lingala is denoted by use of the connective ya followed by the possessor (this takes the translation "the X of Y"). It is also used quite frequently for the object of prepositional phrases that describe a noun, typically some sort of spatial qualifier.
 +
 +
The following examples show the use of nmod for possession:
 +
<pre>
 +
"<mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<amibwakaki>"
 +
"kobwáka" v refl urp p3 sg aa @root #2->0
 +
"<likolo>"
 +
"likoló" n cl5 sg @case #3->5
 +
"<lya>"
 +
"lya" conn @case #4->5
 +
"<mbeto>"
 +
"mbeto" n cl9 sg @obl #5->2
 +
"<ya>"
 +
"ya" conn @case #6->5
 +
"<ye>"
 +
"yé" prn pers p3 sg aa @nmod #7->5
 +
</pre>
 +
 +
This is translated "And he cast himself upon his bed." Or, more literally, "And he cast himself on top of the bed of his." The personal pronoun ye has an nmod dependence to the bed. (Note that likolo is marked with a case dependence here despite being a noun. Likolo means "the top," but can effectively function as a preposition when used with the connective.)
 +
 +
The following example has two examples of the nmod dependence -- one to denote possession and the other to give a spatial qualification through the use of a preposition.
 +
<pre>
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<esalemaki>"
 +
"kosalema" v urp p3 sp nn @root #2->0
 +
"<ete>"
 +
"ete" cnjsub @mark #3->4
 +
"<azongaki>"
 +
"kozónga" v urp p3 sg aa @ccomp #4->2
 +
"<o>"
 +
"o" pr @case #5->6
 +
"<ndako>"
 +
"ndako" n cl1 sg @obl #6->4
 +
"<ya>"
 +
"ya" conn @case #7->8
 +
"<yemei>"
 +
"yemei" prn pers p1 sg aa @nmod #8->6
 +
"<o>"
 +
"o" pr @case #9->10
 +
"<Yelusaleme>"
 +
"Yelusaleme" noun prop @nmod #10->6
 +
"<;>"
 +
";" sent @punct #11->2
 +
</pre>
 +
This is translated "And it came to pass that he returned to his own house at Jerusalem". The nmod dependence for the pronoun yemei indicates his possession of the home and the nmod dependence for the proper noun Yelusaleme indicates the home's location in Jerusalem.
 +
 +
===obl===
 +
As described in the UD documentation, the obl relation is used for nominals that modify verbs, adjectives, or adverbs as a non-core argument. Just as in English, its most typically relates a verb to the object of a preposition following it. For one example, consider the last example of the previous section. "ndako" (house) has an obl dependence to "azongaki" (he returned). The relevant portion of the sentence is translated "he returned to (his own) house."
 +
 +
For a second example, consider the following sentence:
 +
<pre>
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<esalemaki>"
 +
"kosalema" v urp p3 sp nn @root #2->0
 +
"<ete>"
 +
"ete" cnjsub @mark #3->8
 +
"<bo>"
 +
"bó" cnjsub @mark #4->5
 +
"<etangaki>"
 +
"kotánga" v urp p3 sp nn @advcl #5->8
 +
"<ye>"
 +
"yé" prn pers p3 sg aa @nsubj #6->5
 +
"<,>"
 +
"," cm @punct #7->5
 +
"<atondisamaki>"
 +
"kotónda" v caus pass urp p3 sg aa @ccomp #8->2
 +
"<na>"
 +
"na" pr @case #9->10
 +
"<Molimo>"
 +
"molimo" n cl3 sg @obl #10->8
 +
"<mwa>"
 +
"mwa" conn @case #11->12
 +
"<Nkolo>"
 +
"Nkolo" n sg cl1 @nmod #12->10
 +
"<.>"
 +
"." sent @punct #13->2
 +
</pre>
 +
 +
Here, "Molimo" (Spirit) has an obl dependency to atondisamaki (he was filled). The above excerpt is translated "And it came to pass that as he read he was filled with the Spirit of the Lord."
 +
 +
===advcl===
 +
As explained in the UD documentation, an advcl relation is used for any clause that modifies a verb, adjective, or adverb (or other predicate) but is not itself core. In the annotated corpus, these typically modify verbs. For one example, consider the last example of the previous section where "etangaki" (read) has an advcl dependency on  atondisamaki (he was filled). This is because the clause "as he read" describes how he was filled.
 +
 +
For a second example, consider the following partial sentence:
 +
<pre>
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->20
 +
"<emonoki>"
 +
"komona" v urp p3 sg nn @advcl #2->20
 +
"<tata>"
 +
""tata" n cl1 sg @nsubj #3->2
 +
"<wa>"
 +
"wa" pr @case #4->5
 +
"<ngai>"
 +
"ngáí" prn pers p1 sg aa @nmod #5->3
 +
"<ete>"
 +
"ete" cnjsub @mark #6->11
 +
"<mai>"
 +
"maí" n cl5 pl @nsubj #7->11
 +
"<ma>"
 +
"ma" pr @case #8->9
 +
"<ebale>"
 +
"ebale" n cl7 sg @nmod #9->7
 +
"<mazalaki>"
 +
"kozala" v aux p3 pl nn @aux #10->11
 +
"<kosopana>"
 +
"kosopa" v recip inf @ccomp #11->2
 +
"<o>"
 +
"o" pr @case #12->13
 +
"<kati>"
 +
"káti" pr @obl #13->11
 +
"<ya>"
 +
"ya" conn @case #14->15
 +
"<liziba>"
 +
"liziba" n cl5 sg @nmod #15->13
 +
"<lya>"
 +
"lya" conn @case #16->17
 +
"<Mbu>"
 +
"mbu" n cl9 sg @nmod #17->15
 +
"<Etane>"
 +
"kotána" v pres p3 sg nn @compound #18->17
 +
"<,>"
 +
"," cm @punct #19->2
 +
"<alobaki>"
 +
"koloba" v urp p3 sg aa @root #20->0
 +
</pre>
 +
 +
Here, "emonoki" (saw) had an advcl dependency on alobaki (he said). The excerpt is translated "And when my father saw that the waters of the river emptied into the fountain of the Red Sea, he said," so it is seen that the the father seeing the waters of the river emptied into the fountain of the Red Sea describes spoke.
 +
 +
===obj===
 +
The obj relation is quite simple. It is simply used to denote that the object of a verb depends on the verb. Note that the obl relation (described above) is used when a nominal describes the verb in some way but is not directly an object (like a preposition. For one example, consider the following phrase:
 +
<pre>
 +
"<Nkolo>"
 +
"Nkolo" n prop @nsubj #17->18
 +
"<atindaki>"
 +
"kotinda" v urp p3 sg aa @advcl #18->15
 +
"<ye>"
 +
"yé" prn pers p3 sg aa @obj #19->18
 +
</pre>
 +
The above is translated "The Lord commanded him." Clearly, "ye" (him) is the object of "atindaki" ([he] commanded).
 +
 +
For a second example, consider the following phrase:
 +
<pre>
 +
 +
"<Mpe>"
 +
"mpé" cngcoo @cc #1->2
 +
"<amonoki>"
 +
"komóna" v urp p3 sg aa @root #2->0
 +
"<lisusu>"
 +
"lisúsu" adv @advmod #3->2
 +
"<zomi>"
 +
"zomi" num card @compound #4->5
 +
"<na>"
 +
"na" cngcoo @compound #5->6
 +
"<babale>"
 +
"mibale" num card @nummod #6->7
 +
"<basusu>"
 +
"mosusu" n cl1 pl @obj #7->2
 +
</pre>
 +
The above is translated "And he also saw 12 others..." Clearly, "basusu" (others) is the object of "amonoki" (he saw).
  
 
[[Category: sp17_UD]]
 
[[Category: sp17_UD]]
 
[[Category: Lingala]]
 
[[Category: Lingala]]

Latest revision as of 18:01, 24 April 2017

Lingala Universal Dependencies are documented on this page. Related models and corpora are available on GitHub.

Evaluation

UD Parser Evaluation
lin.withmorph.udpipe lin.nomorph.udpipe Number of forms Number of sentences
lin.annotated.ud.conllu UAS: 69.85%, LAS: 61.45% UAS: 61.83%, LAS: 53.44% 262 12
lin.annotated2.ud.conllu UAS: 39.82%, LAS: 21.24% UAS: 29.20%, LAS: 20.35% 113 8

The lower stats for lin.annotated2.ud.conllu are to be expected since the text for this corpus was drawn from a completely different source with a very different style.

Dependency Relations

Examples of dependency relations are shown below. Note that Lingala has some odd constructions with nouns and verbs that can allow them to function as if they were a different part of speech, so sometimes a certain dependency might seem to correspond to an otherwise incompatible part of speech.

ccomp

The ccomp relation is used in the manner described in the UD documentation: it denotes a clause relating to a main verb that is a core argument. This comes up very frequently in the Book of Mormon corpus that was being used for annotations, since the construction "It came to pass that..." is used quite frequently. The verb following "that" has a ccomp dependency to "came."

For a basic example, consider the following partial sentence:

"<Mpe>"
	"mpé" cngcoo @cc #1->2
"<esalemaki>"
	"kosalema" v urp p3 sp nn @root #2->0
"<ete>"
	"ete" cnjsub @mark #3->4
"<azongaki>"
	"kozónga" v urp p3 sg aa @ccomp #4->2

This is translated "It came to pass that he returned..." Here, "returned" has a ccomp dependency to "came." For a slightly more complicated example, consider the following excerpt:

"<Mpe>"
	"mpé" cngcoo @cc #1->2
"<esalemaki>"
	"kosalema" v urp p3 sp nn @root #2->0
"<ete>"
	"ete" cnjsub @mark #3->8
"<bo>"
	"bó" cnjsub @mark #4->5
"<etangaki>"
	"kotánga" v urp p3 sp nn @advcl #5->8
"<ye>"
	"yé" prn pers p3 sg aa @nsubj #6->5
"<,>"
	"," cm @punct #7->5
"<atondisamaki>"
	"kotónda" v caus pass urp p3 sg aa @ccomp #8->2

This is translated "And it came to pass that as he read he was filled..." As before, "filled" has a ccomp dependency to "came," but in this example there is an intervening advcl.

nmod

The nmod relation is used quite frequently. Most frequently, it indicates possession, since possession in Lingala is denoted by use of the connective ya followed by the possessor (this takes the translation "the X of Y"). It is also used quite frequently for the object of prepositional phrases that describe a noun, typically some sort of spatial qualifier.

The following examples show the use of nmod for possession:

"<mpe>"
	"mpé" cngcoo @cc #1->2
"<amibwakaki>"
	"kobwáka" v refl urp p3 sg aa @root #2->0
"<likolo>"
	"likoló" n cl5 sg @case #3->5
"<lya>"
	"lya" conn @case #4->5
"<mbeto>"
	"mbeto" n cl9 sg @obl #5->2
"<ya>"
	"ya" conn @case #6->5
"<ye>"
	"yé" prn pers p3 sg aa @nmod #7->5

This is translated "And he cast himself upon his bed." Or, more literally, "And he cast himself on top of the bed of his." The personal pronoun ye has an nmod dependence to the bed. (Note that likolo is marked with a case dependence here despite being a noun. Likolo means "the top," but can effectively function as a preposition when used with the connective.)

The following example has two examples of the nmod dependence -- one to denote possession and the other to give a spatial qualification through the use of a preposition.

"<Mpe>"
	"mpé" cngcoo @cc #1->2
"<esalemaki>"
	"kosalema" v urp p3 sp nn @root #2->0
"<ete>"
	"ete" cnjsub @mark #3->4
"<azongaki>"
	"kozónga" v urp p3 sg aa @ccomp #4->2
"<o>"
	"o" pr @case #5->6
"<ndako>"
	"ndako" n cl1 sg @obl #6->4
"<ya>"
	"ya" conn @case #7->8
"<yemei>"
	"yemei" prn pers p1 sg aa @nmod #8->6
"<o>"
	"o" pr @case #9->10
"<Yelusaleme>"
	"Yelusaleme" noun prop @nmod #10->6
"<;>"
	";" sent @punct #11->2

This is translated "And it came to pass that he returned to his own house at Jerusalem". The nmod dependence for the pronoun yemei indicates his possession of the home and the nmod dependence for the proper noun Yelusaleme indicates the home's location in Jerusalem.

obl

As described in the UD documentation, the obl relation is used for nominals that modify verbs, adjectives, or adverbs as a non-core argument. Just as in English, its most typically relates a verb to the object of a preposition following it. For one example, consider the last example of the previous section. "ndako" (house) has an obl dependence to "azongaki" (he returned). The relevant portion of the sentence is translated "he returned to (his own) house."

For a second example, consider the following sentence:

"<Mpe>"
	"mpé" cngcoo @cc #1->2
"<esalemaki>"
	"kosalema" v urp p3 sp nn @root #2->0
"<ete>"
	"ete" cnjsub @mark #3->8
"<bo>"
	"bó" cnjsub @mark #4->5
"<etangaki>"
	"kotánga" v urp p3 sp nn @advcl #5->8
"<ye>"
	"yé" prn pers p3 sg aa @nsubj #6->5
"<,>"
	"," cm @punct #7->5
"<atondisamaki>"
	"kotónda" v caus pass urp p3 sg aa @ccomp #8->2
"<na>"
	"na" pr @case #9->10
"<Molimo>"
	"molimo" n cl3 sg @obl #10->8
"<mwa>"
	"mwa" conn @case #11->12
"<Nkolo>"
	"Nkolo" n sg cl1 @nmod #12->10
"<.>"
	"." sent @punct #13->2

Here, "Molimo" (Spirit) has an obl dependency to atondisamaki (he was filled). The above excerpt is translated "And it came to pass that as he read he was filled with the Spirit of the Lord."

advcl

As explained in the UD documentation, an advcl relation is used for any clause that modifies a verb, adjective, or adverb (or other predicate) but is not itself core. In the annotated corpus, these typically modify verbs. For one example, consider the last example of the previous section where "etangaki" (read) has an advcl dependency on atondisamaki (he was filled). This is because the clause "as he read" describes how he was filled.

For a second example, consider the following partial sentence:

"<Mpe>"
	"mpé" cngcoo @cc #1->20
"<emonoki>"
	"komona" v urp p3 sg nn @advcl #2->20
"<tata>"
	""tata" n cl1 sg @nsubj #3->2
"<wa>"
	"wa" pr @case #4->5
"<ngai>"
	"ngáí" prn pers p1 sg aa @nmod #5->3
"<ete>"
	"ete" cnjsub @mark #6->11
"<mai>"
	"maí" n cl5 pl @nsubj #7->11
"<ma>"
	"ma" pr @case #8->9
"<ebale>"
	"ebale" n cl7 sg @nmod #9->7
"<mazalaki>"
	"kozala" v aux p3 pl nn @aux #10->11
"<kosopana>"
	"kosopa" v recip inf @ccomp #11->2
"<o>"
	"o" pr @case #12->13
"<kati>"
	"káti" pr @obl #13->11
"<ya>"
	"ya" conn @case #14->15
"<liziba>"
	"liziba" n cl5 sg @nmod #15->13
"<lya>"
	"lya" conn @case #16->17
"<Mbu>"
	"mbu" n cl9 sg @nmod #17->15
"<Etane>"
	"kotána" v pres p3 sg nn @compound #18->17
"<,>"
	"," cm @punct #19->2
"<alobaki>"
	"koloba" v urp p3 sg aa @root #20->0

Here, "emonoki" (saw) had an advcl dependency on alobaki (he said). The excerpt is translated "And when my father saw that the waters of the river emptied into the fountain of the Red Sea, he said," so it is seen that the the father seeing the waters of the river emptied into the fountain of the Red Sea describes spoke.

obj

The obj relation is quite simple. It is simply used to denote that the object of a verb depends on the verb. Note that the obl relation (described above) is used when a nominal describes the verb in some way but is not directly an object (like a preposition. For one example, consider the following phrase:

"<Nkolo>"
	"Nkolo" n prop @nsubj #17->18
"<atindaki>"
	"kotinda" v urp p3 sg aa @advcl #18->15
"<ye>"
	"yé" prn pers p3 sg aa @obj #19->18

The above is translated "The Lord commanded him." Clearly, "ye" (him) is the object of "atindaki" ([he] commanded).

For a second example, consider the following phrase:


"<Mpe>"
	"mpé" cngcoo @cc #1->2
"<amonoki>"
	"komóna" v urp p3 sg aa @root #2->0
"<lisusu>"
	"lisúsu" adv @advmod #3->2
"<zomi>"
	"zomi" num card @compound #4->5
"<na>"
	"na" cngcoo @compound #5->6
"<babale>"
	"mibale" num card @nummod #6->7
"<basusu>"
	"mosusu" n cl1 pl @obj #7->2

The above is translated "And he also saw 12 others..." Clearly, "basusu" (others) is the object of "amonoki" (he saw).