A colloquium presented by Joseba Abaitua (Universidad de Deusto in Bilbao, Spain) at the Computer Research Laboratory, New Mexico State University, Las Cruces, New Mexico. July 1998.

Processing
inflectional head initial
vs.
agglutinative head final
languages in an Spanish-Basque bitext

[Lesson 1] [Lesson 2] [Lesson 3] [Lesson 4] [Lesson 5] [Lesson 6] [Lesson 7]
[References]

Lesson 1

Aldapeko sagarraren
adarraren puntan
puntaren puntan
txoria zegoen kantari
txiruliruli, txiruliruli
nork dantzatuko ote du
soinutxo hori?

Mini-lexicon for Basque

[up] [next]

Lesson 2

The structure of Basque phrases

Basque vs. Spanish phrases

Figure 1 Head final vs. head initial phrases

[up] [next]

Lesson 3

Mini-grammar for Basque PPs:

See note 1.

[up] [next]

Lesson 4

Mini-grammar for Basque NPs:

See note 2.

Category order under Basque NP: S'(rel) > AP > No > Dgr > N > A > Det

Category order under Spanish NP: Det > No > N > Dgr > A > PP > S'(rel)

The categories S', AP and PP above are recursive. Therefore, contrary to Spanish, Basque, as Japanese, suffers from left recursion, which is a well-know problem for straight top-down parsers (DCGs, etc.). Japanese researches realized very quickly and developed bottom-up parsers to solve the problem (such as BUP, by Matsumoto et al. 83, which was later applied to Basque by Ruiz et al. 91).

[up] [next]

Lesson 5

Grammatical suffixes in Basque:

See note 3.

Complements of the Noun in Basque are derived by means of two suffixes. Complements derived from Noun Phrases take the suffex ren and Postpositional Phrases take ko. Given that these complements show conceptual and situational similarities with some adjectives (eg. euskal 'Basque'), we use the category AP (Adjectival Phrase) to define them.

Tree representation

Figure 2 Tree structure of phrase in (2)

[up] [next]

Lesson 6

Agglutinative suffixation in Basque:

This elliptical derived nominal has several anaphoric elements. We can think of it as an answer to a sentence as:

Zein sagarren adarri zaude begira? Aldapekoarenari.
¿A la rama de qué manzano miras? A la del de la cuesta.
To the branch of which appletree are you looking at? To the one of the one on the hillside.

[up] [next]

Lesson 7

Processing Spanish and Basque phrases in a bitext

[The bitext: a sample] [Techniques] [RBBP] [HBP] [Large language resources]

The bitext

• Since 1980 all official publications must be bilingual in the Basque Country of Spain.
• 300 translators: 70,000 pages
• 80% of Spanish > Basque published translations (remaining 20% is literature, education and media)
• Yet, according to IVAP, only 20% of official documentation is translated
• Main institutions:
  • Basque Government (BOPV)
  • Basque Parlament
  • County Councils (Diputaciones) of Alava, Bizkaia, and Gipuzkoa (BOA, BOB, BOG)
• Our corpus (aprox. 7 million words)
  • BOB (1991-1995)
  • BOA (1989-1993)
  • BOPV (1995)
• The bitext (aprox. 500,000 words), only document type Orden Foral from BOB.

[up] [next]

A bitext sample:

[up] [next]

Techniques for bitext processing:

• Rule based processing
• Hybrid techniques
• How bitexts become large language resources

Rule based processing

[up] [next]

Hybrid techniques

A combination of techniques (structural and POS tagging, heuristics, cognate matching, dictionary lookup, statistics, rules, etc. see Martínez et al. 98) have been used to process the bitext. This has resulted in the segmentation and alignment of:

• Proper names
• Terminological units
• Word co-occurrences
• Sentences
• Paragraphs
• Text divisions
• Documents

Example 9. Segmented translation units

[up] [next]

How bitexts become large language resources

An aligned bitext encoded by means of rich descriptive markup becomes a large language resource in itself (we follow TEI P3 guidelines for SGML markup). But the following additional resources can also be derived, as a by-product, from the annotated and aligned bitext:

• Three translation memory bases
  • TMbase1, made of aligned <s>
  • TMbase2, made of aligned DTD elements: <opener>, <div>, <seg>, etc.
  • TMbase3, made of aligned documents.
• Two bilingual glossaries
  • A termbase, containing aligned <term>
  • Proper name collections, aligned <rs>
• Aligned DTDs (document grammars)

[up] [next]

Notes 1, 2 and 3.

1. Things are more complicated than represented here. Itziar Laka has a good introductory Basque grammar that can be consulted on the Internet.

Both Basque cases and postpositions have plural as well as singular forms (adarra[abs sg]/adarrak[abs pl]; adarrak[erg sg]/adarrek[erg pl]; adarrari[dat sg]/adarrei[dat pl], etc.).

It must be pointed out also that a large number of morphophonological rules apply when cases, postpositions and suffixes get agglutinated. These rules will depend on such things as whether the nominal stem ends in a vowel or a consonant, whether two vowels are together, and so on: adar-a > adarra, aldapa-a > aldapa, punta-ei > puntei, etc.

In any case, these rules will be interpreted differently depending on the analyst's affiliation. Linguists will be more concerned about etymological and morphophonologycal issues. Programmers will try to cope with data in the most implementable way, whatever the etymological reality of their rules may be. We aggree with the former, but write our rules largely as the latter. Two further explanations are due here:

2. Our NP rules as explained here might be somehow contentious. The agglutination of determiner and case suffixes makes them become compacted and appear more like inflections than isolated suffixes (in a sense, Basque is both agglutinative and inflectional, as is seen also in finite verbs). It is therefore difficult to truely separate grammatical suffixes clearly, although it is very convenient for the sake of writing phrase structure grammar rules. We will mantain our NP rules as:

3. Laka explains the difference between the ko and ren suffixes in the following manner: "the morpheme ko can indicate location, and this is why it is sometimes referred to as a 'locative genitive', but location is not the only relation it can convey. However, one general guideline that is helpful in distinguishing the use of ko and ren phrases involves location: ko is attached to phrases that denote location, or phrases that denote a property. All other relations a phrase may bear with respect to a Noun are dealt with by means of the morpheme ren. " (See more examples in Laka's section on Complements of [Basque] Nouns).

One major feature of the ko suffix, related to the locative denotation, is that it is used to derive adjectival phrases from postpositional phrases. The latter are not allowed inside noun phrases (unless they get a ko attached). This fact permits the generalization expressed by the rule AP -> PP -ko.

The ko suffix has however a wider derivational force. There are also cases involving participal clauses, which are a much used alternative to relative clauses:

Ean erosi genituen hiru sagar gorriak
Ea-in buy have three apple red-det
The three red apples that we bought in Ea
Example 11

Ean erositako hiru sagar gorriak
Ea-in buy-ta-ko three apple red-det
The three red apples bought in Ea
Example 12

The -ta ending is used to give participles a resultative meaning.

But on top of PP and participles, ko can also be used to derive predicative phrases out of nominal constructions:

Sapore oneko sagarra
Taste good-ko apple-det
An apple with good taste
Example 13

There has been some discussion concerning the nature of the underlying phrase that is attached to the ko suffix. However things may be, for our brief introuduction we will asume that, at least for phrases involving location, there does exist an underlying PP:

As is the case of example (2) above: aldapa-a-n-ko > aldapeko

Although Basque morphology is complex to deal with, it has been successfully processed by means of a two-level morphological analyzer. See the references to the sofware developed by the IXA Group at the Euskal Herriko Unibertsitatea.

[up]

References

• Joseba Abaitua, Arantza Casillas, Raquel Martínez, Legebiduna project
• Eneko Agirre, Iñaki Alegria, Xabier Artola, Arantza Díaz de Ilarraza, Montserrat Maritxalar, Kepa Sarasola, Mirian Urkia (IXA Group), Xuxen: Spelling Checker/Corrector for Basque based in Two-Level Morphology. Proceedings of ANLP'92, 119-125. Povo Trento. 1992.
• Computer Research Laboratory, http://crl.nmsu.edu
• Euskal Herriko Unibertsitatea, http://www.ehu.es/
• IXA Group, http://ixa.si.ehu.es/
• IXA Group, List of papers on Morphology.
• IXA Group, MORFEUS, A Morphological Analyser for Basque
• ITEM Project, http://sensei.ieec.uned.es/item/
• Itziar Laka. A Brief Grammar of Euskara, the Basque Language (***) (a very handy grammar for linguists )
• Itziar Laka. Complements of the [Basque] Noun
• Instituto Vasco para la Administración Pública, http://www.ivap.es
• Raquel Martínez, Joseba Abaitua, Arantza Casillas. Aligning tagged bitexts. Workshop on Very Large Corpora, COLING/ACL'98. Montreal . 1998.
• Yuji Matsumoto, Hozumi Tanaka, H. Hirawaka, Hideo Miyosi, Hideki Yasukawa. BUP: a bottom-up parser embedded in PROLOG, New Generation Computing, 145-158, Vol. 1, 1983.
• New Mexico State University, http://www.nmsu.edu
• Juan Carlos Ruiz, Juan Ramón Zubizarreta, Joseba Abaitua, GFU-LAB: entorno de desarrollo de gramáticas y su aplicacion al euskara, Procesamiento del Lenguaje Natural, 1991.
• Universidad de Deusto, http://www.deusto.es/

Other Basque pages:

• Iñaki Agirre, Basque Language Course (Basque lessons twice a week through a BASQUE-L mail list)
• Eugenio Alberdi, Basque Language Page (links)
• Josu Aranbarri, Ibon Basterretxea, Euskara, The Language of Basque People (brief grammar, vocabulary)
• Jon Aske, Basque Language Page (proverbs, article archive, links)
• ComStar Company, Basque Software Digest (interesting but confusing and incomplete)
• Alan R. King, Introduction to Basque (downloadable MS-Word 66-page manual)
• Larry Trask, Basque Page (***) (FAQs about Basque)
• Blas Uberuaga, Introduction to Euskara (***) (verb charts, proverbs, sports, cuisine, links, etc.)

[up]