Upload
dawn-price
View
48
Download
0
Embed Size (px)
DESCRIPTION
Systematic Parameterized Description of Pro-forms in the Prague Dependency Treebank 2.0. Magda Ševčíková Zdeněk Žabokrtský Institute of Formal and Applied Linguistics Charles University Prague, Czech Republic {sevcikova,zabokrtsky}@ufal.mff.cuni.cz. Outline of the talk. Introduction - PowerPoint PPT Presentation
Citation preview
Systematic Parameterized Description of Pro-forms in the Prague Dependency
Treebank 2.0
Magda Ševčíková
Zdeněk Žabokrtský
Institute of Formal and Applied Linguistics
Charles University
Prague, Czech Republic
{sevcikova,zabokrtsky}@ufal.mff.cuni.cz
TLT 2006 [email protected]/20
Outline of the talk
Introduction
Description of pro-forms in the PDT 2.0 Type 1
• Personal pronouns
Type 2
• Indefinite, negative, interrogative, and relative pronouns
• Pro-adverbs and pro-numerals
Pro-forms in other languages
Final remarks
TLT 2006 [email protected]/20
Introduction Pro-forms
pronouns, pro-adverbs, and pro-numerals closed classes to replace or substitute other words, phrases, or sentences anaphoric and deictic functions semantically relevant regularities within the sub-classes
• nobody-never-nowhere• everybody-always-everywhere
Pro-forms in the PDT 2.0 formal linguistic system for annotation of pro-forms
making the present regularities explicit part of the deep-syntactic layer (tectogrammatical layer, t-layer) representation by a reduced set of (underlying) lemmas in combination with
relevant attributes
TLT 2006 [email protected]/20
PDT project Historical background
mid 1960’s Functional Generative Description (Petr Sgall et al.)
1994 Czech National Corpus 1995 PDT started 1998 PDT 0.5 pre-release 2001 PDT 1.0 released by LDC (LDC2001T10)
manual annotation of morphology and surface syntax
2006 PDT 2.0 released by LDC (LDC2006T01) interlinked morphological, surface-syntactic
and complex deep-syntactic annotation
TLT 2006 [email protected]/20
PDT 2.0 Layers of annotation
Lit: [He] was would went to forest.[He] would have gone to the forest.
Tectogrammatical layer deep-syntactic dependency tree 59 % of the a-layer data 3,165 doc., 49,431 sent., 833,195 tokens
Analytical layer surface-syntactic dependency tree 75 % of the m-layer data 5,330 doc., 87,913 sent., 1,503,739 tokens
Morphological layer m-lemma and m-tag
associated with each token 7,110 textual documents 115,844 sent., 1,957,247 tokens
Word layer original text, segmented on word boundaries
TLT 2006 [email protected]/20
Outline of the talk
Introduction
Description of pro-forms in the PDT 2.0 Type 1
• Personal pronouns
Type 2
• Indefinite, negative, interrogative, and relative pronouns
• Pro-adverbs and pro-numerals
Pro-forms in other languages
Final remarks
TLT 2006 [email protected]/20
Description of pro-forms in the PDT 2.0
M-layer pronouns, pro-adverbs, and pro-numerals treated separately m-lemma, m-tag
T-layer 2 basic types of description
• type 1: personal pronouns
• type 2: indefinite, negative, interrogative, and relative pronouns together with pro-adverbs and pro-numerals
semantic features originally present in the word form extracted and stored as values of inner attributes of the t-node that corresponds to the given word form
TLT 2006 [email protected]/20
TLT 2006 [email protected]/20
Type 1Personal pronouns in the PDT 2.0
all personal pronouns (no matter whether they are pro-dropped or present in the sentence) represented by nodes labeled with a single, artificial lemma #PersPron
grammatical information expressed by a personal pronoun in the sentence is stored in node attributes person, number, and gender
attribute politeness for discerning between honorific and non-honorific usage
vy jste přišel (you came said politely to a single person) #PersPron + 2nd person + singular + masc.anim. + polite
TLT 2006 [email protected]/20
Tím, že Evropská unie nechala ve rwandské operaci Francii na holičkách, podle Léotarda ukázala, že její politika nemá žádný africký rozměr. According to Léotard, by the fact that the European Union left France in the lurch concerning the Rwanda operation, [it] has shown that its politics has no African dimension.
at the t-layer, representation of personal pronouns was completed with the annotation of co-reference (i.e relations between nodes referring to the same entity)
Type 1Personal pronouns and co-reference
TLT 2006 [email protected]/20
Type 2Indefinite, negative, interrogative, and relative pronouns in the PDT 2.0 in Czech, single meanings are expressed regularly by means of a relatively
small group of prefixes that join together with a small set of bases transparent correspondence between the semantic features and formal
composition of pronouns: indefinite prefix ně-: někdo (somebody) – něco (something) – nějaký (some) negative prefix ni-: nikdo (nobody) – nic (nothing)…
at the t-layer, pronouns with the same base element grouped together, each pronoun group represented by the lemma corresponding to the respective relative pronoun: e.g. někdo (somebody) and nikdo (nobody) represented by the lemma kdo (who)
corresponding possessive pronouns represented in the same way as the non-possessive ones
the semantic feature completing the reduced lemma was stored in the indeftype attribute
TLT 2006 [email protected]/20
Type 2Indefinite, negative, interrogative, and relative pronouns and the indeftype attribute
all indefinite, negative, interrogative, and relative pronouns represented by only four lemmas at the t-layer
the reduced lemmas were completed by a value of the indeftype attribute 11 values:
TLT 2006 [email protected]/20
Type 2Pro-adverbs and pro-numerals in the PDT 2.0
in Czech, pro-adverbs (e.g. nikde (nowhere), nějak (somehow)) and pro-numerals (e.g. několik (a few)) share certain semantic features with pronouns
represented in the same way as indefinite, negative, interrogative, and relative pronouns at the t-layer
another derivational relation can be seen between pro-adverbs with directional meaning and those of location – for example, the adverb odněkud (from somewhere) is represented as follows: lemma kde (where) + indef1 value (of the indeftype attribute) + functor
DIR1 capturing the directional meaning
TLT 2006 [email protected]/20
Zakládá-li si někdo na tom, že se vyhýbá cizím slovům, pak udělá nejlíp, když se nikdy nepodívá do Etymologického slovníku jazyka českého. If someone finds it important that [he] eliminates foreign words, then the best thing [he] can do is if [he] never looks in the Etymology Dictionary of Czech.
TLT 2006 [email protected]/20
Outline of the talk
Introduction
Description of pro-forms in the PDT 2.0 Type 1
• Personal pronouns
Type 2
• Indefinite, negative, interrogative, and relative pronouns
• Pro-adverbs and pro-numerals
Pro-forms in other languages
Final remarks
TLT 2006 [email protected]/20
Pro-forms in other languagesPDT-like description indefinite, negative, interrogative, and relative pronouns and other pro-
forms are unproductive classes with (at least to a certain extent) transparent derivational relations also in other languages
preliminary sketch of several English and German pronouns:
still not solved: English anybody, German niemand and nirgendjemand …
TLT 2006 [email protected]/20
Lit.: The teacher finds nowhere a mistake.
Der Lehrer findet nirgends einen Fehler.
In Helbig, H. (2001), Die semantische Struktur natürlicher Sprache, Springer, 2001, p. 174
Negative pro-adverbs
Lit.: Peter goes on holiday nowhere.
Peter fährt in den Ferien nirgendwo hin.
with directional meaning with local meaning
Pro-forms in other languagesHelbig’s MultiNet
TLT 2006 [email protected]/20
Outline of the talk
Introduction
Description of pro-forms in the PDT 2.0 Type 1
• Personal pronouns
Type 2
• Indefinite, negative, interrogative, and relative pronouns
• Pro-adverbs and pro-numerals
Pro-forms in other languages
Final remarks
TLT 2006 [email protected]/20
Final remarks achievements:
all pro-forms in Czech divided into two groups:• personal (and corresponding possessive) pronouns• indefinite, negative, interrogative, and relative pronouns (and
corresponding possessive pronouns) and pro-adverbs and pro-numerals several pro-form analogies crossing the part-of-speech boundaries are
explicitly marked in the annotation verification of the formal system on large-scale data
future work: to elaborate the system for other languages in more detail, taking into
consideration specific phenomena of the respective language to describe the relations among pro-form systems in more languages
(for example, for the purposes of machine translation)
TLT 2006 [email protected]/20
http://ufal.mff.cuni.cz/pdt2.0/