Ros Kat presentation

Transfer rules in Bulgarian-English Machine TranslationThe POS (Noun) and the Noun PhraseRossitsa Petcova, Katya AlahverdzhievaSofia University: 

Transfer rules in Bulgarian-English Machine Translation The POS (Noun) and the Noun Phrase Rossitsa Petcova, Katya Alahverdzhieva Sofia University

Main MT approaches:: 

Main MT approaches: Example-based vs. Rule-based Approaches Example-based or memory-based – an analogy-based translation system has pairs of bilingual expressions stored in an example database. The source language input expression is matched against the source language examples in the database, and the best match is chosen. The system returns the TL equivalent of this example as output. Disadvantages: the database becomes too large and unmanageable for systems with a realistic coverage. Solution: a combination of the analogy-based approach with the rule-based approach, hybrid systems.

Rule-based methods: 

Rule-based methods

Interlingua method: 

Interlingua method Main features of Interlingua: provides an abstract representation of the meaning a universal representation Independent from the SL Good for multilingual systems - easy to add new languages Drawback: difficult for representation, How to build a really language-neutral meaning representation? Steps in translation using Interlingua: 1) translate source sentence into meaning representation (Interlingua) 2) generate target sentence from meaning Source text --[analysis]-- Interlingua --[synthesis]-- Target text Bulgarian --[analysis]-- Interlingua --[synthesis]-- English English --[analysis]-- Interlingua --[synthesis]-- Bulgarian

Transfer method:: 

Transfer method: Based on transfer rules – rules mapping the surface structure of the two languages Language specific - rules reflect the structure of the language in question Unidirectional – you need to have different sets of transfer rules for the two direction of translation Works in 3 stages: Source text --[analysis]– Intermediate Structure (source) --[transfer]– Intermediate Structure (target) --[synthesis]-- Target text Bg text – analysis – Int Str - Bulgarian-English Transfer Rules– Int Str – synthesis – English text English text –analysis - Int Str - English-Bulgarian Transfer Rules– Int Str – synthesis – Bg text

The transfer method – knowledge base: 

The transfer method – knowledge base 3 stages: Analysis of SL- Transfer - Generation of TL Required Knowledge base for the transfer approach: Description of the grammar of the SL Description of the grammar of the TL Lexicon of the SL Lexicon of the TL Bilingual lexicon Transfer rules

How to write transfer rules for POS : 

How to write transfer rules for POS To define the POS for SL and TL – N, V, Adj, etc Are they variable or invariable If variable, what features do they have in each language– number, person, definiteness, tense, mood, etc

POS in Bulgarian and English: 

POS in Bulgarian and English POS – Bulgarian Variable POS: Noun Verb Pronoun Adjective Numeral Invariable POS: Adverb Preposition Conjunction Interjection POS – English Variable POS: Noun Verb Pronoun Adjective – degrees of comparison Invariable POS: Adverb Preposition Conjunction Interjection Numeral

Features of the NOUN: 

Features of the NOUN Noun in Bulgarian N_Gender = m + f + n; N_Number = s + p; N_Definiteness = 0 + d + h + l; N_Countform = c; N_Vocative case = v; N_Sem = Conc + Abstr + Hum + Mass; Noun in English none N_Number = s + p; DET_Definiteness = 0 + indef + def; none none N_Sem = Conc + Abstr + Hum + Mass;

POS - features: 

POS - features BULGARIAN Pronouns: PRO_Person = 1 + 2 + 3; PRO_Number = s + p; PRO_Gender = f + m + n; Verbs: V_Person = 1 + 2 + 3; V_Number = s + p; V_Mood = i; Adjectives: A_Gender = m + f + n; A_Number = s + p;’ A_Definiteness = 0 +d + sh + l; Numerals: N_Gender = m + f + n; N_Number = s + p;’ N_Definiteness = 0 +d + sh + l; ENGLISH Pronouns: PRO_Person = 1 + 2 + 3; PRO_Number = s + p; PRO_Gender = f + m + n; Verbs: V_Person = 1 + 2 + 3; V_Number = s + p; V_Mood = i; Adjectives: invariable ( forms of comparison) Numerals: invariable

The Noun in Bulgarian - paradigm: 

The Noun in Bulgarian - paradigm Singular, without article N-Sg-0 жена/стол Singular, definite article N-Sg-d жената Singular, short definite article (m, Object) N-Sg-short стола Singular, long definite article (m, Subject) N-Sg-long столът Plural, without article N-Pl-0 жени/столове Plural, definite article N-Pl-d жените Counting form ( m) N-Pl-Count-0 стола Vocative case form ( m , f) Voc жено TOTAL: 8 forms Article: in postposition Masculine Nouns: 7 forms, all forms Feminine Nouns:5 forms, No counting form, no short/long article distinction Neuter Nouns: 4 forms, No vocative, No counting form, no short/long article distinction

English Noun - forms: 

English Noun - forms Singular, no article N-Sg-0 chair Singular, indefinite article N-Sg- Indef a chair Singular, definite article N-Sg-Def the chair Plural, no article N-Pl-0 chairs Plural, definite article N-Pl-Def the chairs Total: 5 forms The Article: a separate item, forms an NP > Det N> a N/ the N

Bulgarian-English Transfer rules for Nouns 1: 

Bulgarian-English Transfer rules for Nouns 1 SINGULAR NOUNS: N-Sg-0 (любов) > Noun-Sg-0 (love) (for abstract notions, uncountable nouns (songs about love) N-Sg-0 (жена) > Noun-Sg- Indef (a woman) (for countable nouns) N-Sg-Def (жената) > Noun-Sg-Def (the woman) N-Sg-Def (котката) > Noun-Sg-Indef (a cat is an animal.) ( speaking in general) N-Sg-Def (Любовта е сляпа.) > Noun-Sg-0 (Love is blind.) (abstract notions) N-Sg-short (работника) > Noun-Sg-Def (the worker) N-Sg-long (работникът) > Noun-Sg-Def (the worker) Vocative (жено) > Noun-Sg-0 (woman)

Bulgarian-English transfer rules for Nouns - 2: 

Bulgarian-English transfer rules for Nouns - 2 Plural nouns N-Pl-0 (столове) > Noun-Pl-0 (chairs) N-Pl-Def (столовете) > Noun-Pl-Def (the chairs) N-Pl-Def (столовете) > Noun-Pl-0 (chairs) (when speaking in general - “Chairs are usually made of wood.” ) N-Pl-Count-0 (стола) > Noun-Pl-O (chairs) (this usage needs context, it is always used with a premodifier denoting quantity, For ex: how many/ three/several/ a few chairs)

Conclusions-1: 

Conclusions-1 Major differences: In Bulgarian – 8 forms , In English – 5 forms Determiners: In Bulgarian – as a suffix, in postposition In English – as a separate item, in preposition, forms a phrase – NP = Det + N Noun-Sg- Indef = Indef (a) +N = Det + N = NP Noun-Sg-Def = Def (the) +N = Det + N = NP Noun-Pl-Def = Def (the) +N = Det + N = NP Rules for NPs: NP::NP [N]> [Det N] жена> a woman жената > the woman

Conclusions-2: 

Conclusions-2 Singular Nouns: In English – no gender distinction , therefore no short/long article distinction for masculine nouns In English-Bulgarian translation of masculine nouns, one has to choose between: Noun-Sg-Def (the worker) > N-Sg-short (работникът) Noun-Sg-Def (the worker) > N-Sg-long (работника) In English - the form for the Vocative case: coincides with the zero-article form Plural Nouns: In English - no distinction in plural for counting forms for masculine nouns: Noun-Pl-0 (chairs) > N-Pl-0 (столове) Noun-Pl-0 (chairs) > N-Pl-Count-0 (стола)

Structural Transfer: 

Structural Transfer Applies to syntactic differences between languages; The source sentence is analyzed to give a tree representing its phrase structure; Transfer rules specify how to translate this tree into a phrase structure tree for the TL; Generate the target sentence from the target tree that has been derived.

Noun Phrase:: 

Noun Phrase: NPs form the majority of textual content of the scientific and technical documents; To survey the differences between Bulgarian and English constituent structure. Phrases that have a NOUN as a headword in preposition {Det, A, Qu, N, Adv} in postposition {N, PP, Adv, Pron poss}

Four major components:: 

Four major components: the determinative: determines the reference of the NP in its linguistic or situational context; premodification: comprises all the modifying or describing constituents before the head, other than determiners; the head: around it the other constituents cluster; and postmodification: comprises all the modifying constituents placed after the head.

Features of the definite article in Bulgarian: 

Features of the definite article in Bulgarian Determiners: Definite article; Demonstrative pronoun. Article: has the phonological properties of an affix, not a clitic.  influenced phonologically by the final vowel of the word – the article attaches, for example, to adjectives and nouns that end in –a, regardless of gender and number (zhena-ta, bashta-ta, deca-ta, etc).  depends on the morphological gender of that word – it attaches to feminine forms regardless of the phonetic final position of the word (esen-ta, prolet-ta).

Phrases with a noun as a headword premodified by an unarticulated adjective : 

Phrases with a noun as a headword premodified by an unarticulated adjective [istoricheska kniga] ‘a history book’ [amerikanska [istoricheska kniga]] ‘an American history book’ [iljustrovana [amerikanska [istoricheska kniga]]] ‘an illustrated American history book’ [naskoro publikuvana [iljustrovana [amerikanska [istoricheska kniga]]]] ‘a recent illustrated American history book’ [protivorechiva [naskoro publikuvana [iljustrovana [amerikanska [istoricheska kniga]]]]] ‘a recent controversial illustrated American history book’ NP::NP [NP[N`ADJ[N`ADJ[N`ADJ N]]]] -> [NP[N`ADJ[N`ADJ[N`ADJ N]]]] NP::NP [ADJ+ N] -> [ADJ+ N]

Phrases with a noun as headword premodified by an articulated adjective : 

Phrases with a noun as headword premodified by an articulated adjective interesna-ta kniga ‘the interesting book’ goljama-ta interesna kniga ‘the big, interesting book’ The article attaches to the right of the first word in the NP. NP::NP [NP[AP[APADJ-Det] ADJ] N]->[NP[AP[APADJ DET] ADJ] N]

Adv + Adj + N: 

Adv + Adj + N When the adjective is preceded by an adverbial modifier, the article must still follow the adjective. It can never simply follow the adverb: mnogo hubavi-te knigi ‘the very nice books’ dosta glupava-ta zabelezhka ‘the quite stupid remark’ priblizitelno sto-te dushi ‘the approximately 100 people’. NP::NP [NP[APADV [APADJ-Det]] N] -> [DPDET [NP[APADV ADJ] N]]

More examples:: 

More examples: [pochti nerazrabotena-ta u nas] problematika [almost not-worked-out-the by us] problematics ‘the problematics (which are) almost not worked out here [poluchena-ta s maka] stipendija [received-the with pain] scholarship ‘the received with pain scholarship’ [verni-jat na demokratichni idei] prezident faithful-the to democratic ideas president ‘the president (who is) faithful to democratic ideas’ [kupeni-te vchera] knigi bought-the yesterday books ‘the books (which were) bought yesterday’ NP::NP [NP[[APADJ-Det] ADV] N] -> [DPDET[NPN[APADJ ADV]]]

Noun as a premodifier: 

Noun as a premodifier chuval kartofi ‘a sack of potatoes’ samun hljab ‘a loaf of bread’; vrazka luk ‘a rope of onions’ vrazka kljuchove ‘a bunch of keys’; jato ptici ‘a flock of birds’; stado kravi ‘a herd of cows’; stado ovce ‘a flock of sheep’; kupa oriz ‘a bowl of rice’ NP::NP [NP N N]->[NPN [PPPrepN]]

Quantifiers as premodifiers: 

Quantifiers as premodifiers Cardinal numbers; Ordinal numbers; All, both, each: free word order Before the headword: Vsichki te pritezhavat zaviden intelekt. ‘Аll of them are extremely intelligent’ After main verbs, an auxiliary, a modal, or “be”, rather than directly after the noun or pronoun: Te vljazoha vsichki zaedno ‘They came in all together’; Pismata bjaha podpisani vsichki. ‘The letters have all been signed’.

Vsichki/each: 

Vsichki/each ‘vsichki’ (each) may appear with a definite article and then the NP before it will also be articulated: knigi-te gi prochetoh vsichki-te ‘the books I read them all’ *knigi-te gi prochetoh hubavi-te ‘*the books I read them nice’ vsichki-(te) tezi knigi ‘all these books’

Phrases modified by adverbs : 

Phrases modified by adverbs rano sutrinta ‘early in the morning’ NP::NP [NP ADJ N] -> [NP ADJ PP]

Apposition: 

Apposition subordinate phrases, expressed by nonprepositional nouns, i.e. we have the line NP NP. They come either in preposition or in postposition: Uncle John, Professor Smith (in preposition); Gorata zakrilnica, the forest protector Valchicata majka, the wolf mother (in postposition)

PP as a Postmodifier: 

PP as a Postmodifier prevod ot italianski na balgarski ezik ‘translation from Italian into Bulgarian’ NP::NP [[NPN[PPPrepN]]PPPrepN] -> [[NPN[PPPrepN]]PPPrepN] recursive order: Vidjah [NP drazhkite [PP za koshnicata [PP s jabylki [PP ot gradinata [PP na saseda [PP ot moja kvartal [PP v centara [PP na grada]]]]]]]. I saw the [NP handles [PP for the basket [PP with the apples [PP from the garden [PP of my neighbour [PP from my neighbourhood [PP in the centre [PP of the town]]]]]]].

Dative possessive clitics : 

Dative possessive clitics Their syntactic position depends directly not on the headword, but on the position of the definite affix, in other words the short possessive clitic appears immediately after the articulated element in the DP, whatever that element happens to be. baba-ta i – grandmother-the her – ‘her grandmother’ umorena-ta i baba – tired-the her grandmother stara-ta i umorena baba –old-the her grandmother – her old tired grandmother

More supporting examples: : 

More supporting examples: [pochti nerazrabotena-ta mu u nas] problematika almost not-worked-out-the him.obl. by us problematics ‘his problematics (which are) almost not worked out here’ [verni-jat ti na demokratichni idei] prezident faithful-the you.obl to democratic ideas president’ ‘your president (who is) faithful to democratic ideas’ [kupenite i vchera] knigi bought-the her.obl yesterday books ‘her books (which were) bougth yesterday’ [zabranena-ta mu ot zakona] kniga fobidden-the me.obl by law book ‘my book which is forbidden by law’

NP as a whole clause: 

NP as a whole clause This is the house that Jack built. This is the malt that lay in the house that Jack built. This is the mouse that ate the malt that lay in the house that Jack built. This is the cat that scared the mouse that ate the malt hat lay in the house that Jack built. This is the dog that chased the cat that scared the mouse that ate the malt that lay in the house that Jack built. This is the boy who loves the dog that chased the cat that scared the mouse that ate the malt that lay in the house that Jack built.

General conclusions:: 

General conclusions: Transfer approach: Detailed analysis of SL and TL grammars Different sets of rules for the two directions of translation English-Bulgarian - more analyzed, existing systems: BULTRA, WebTrance Bulgarian-English - less explored