Difference between revisions of "Norwegian HPSG grammar NorSource"

Revision as of 20:54, 16 April 2014

See the web demo of the Norwegian HPSG grammar Norsource.

License

[WWW] Lesser General Public License For Linguistic Resources

History and Purpose of the grammar

NorSource is a so-called ‘deep’ computational grammar (‘DG’) of Norwegian, developed throughout the last 12 years. The grammar has been developed with a view to the following overall desiderata:

Desideratum 1. Encoding of Linguistic Meaning

As a ‘generic’ information repository, a DG should have a semantic component from which a Reasoning capacity ideally could be deduced for any domain of discourse – possibly with addition of concepts for the specific domains. It should be like a Fregean ‘Sinn’, in acting as a function from domains of use to models of interpretation. However, contrary to most artificial ‘reasoning’ devices, a DG must span the full complexity of a natural language, reflecting the size of its vocabulary and its grammar complexity. In this respect,the DG can also be seen as the materialization of a Generative Grammar, in the original sense of that notion.

Desideratum 2. Cross-grammar Generality

The content of the DG should to as large an extent as possible be phrased in terms used or alignable with terms used in other grammars and for other languages, thereby enabling linguistic comparison using the DG. By ‘content of the DG’ we mean both the content of the grammar files (formalism, notions used) and the content of its parse productions.

Desideratum 3. Interoperability

The DG should attain as much interoperability with other applications as possible. In general, what a digital ubiquitous research environment for linguistics should enable is an interconnectivity of data, researchers and processing facilities whereby from any point in an overall structure of components, a contribution can have its ramifications immediately implemented throughout the entire structure. Such interconnectivity will have to be manifested both on an ‘outer’ level enabling data flow and easy access, and on an ‘inner’ level ensuring information exchange from one system component to another. For a DG, thus, its files and productions (parses, etc.) should be transportable to other applications, and the codes in which its files are written should be readable by other applications, or able to be mapped into other codes.

Desideratum 4. Sustainability

The DG should be in such a format, and be situated in such an over-all environment, that as much as possible of its capacity can be retained, independently of particular persons maintaining it or particular physical environments.

The first of the desiderata reflects a central concern throughout modern logic and philosophy of language, and in turn linguistics and Artificial Intelligence. Semantics being inevitably the basis for significant progress in cross-linguistic modelling, the desideratum has relevance also for desideratum 2. The grammar to be discussed belongs to a family of DGs whose design quite explicitly caters for this concern. This family of DGs has as its formal and theoretical framework HPSG (Pollard and Sag 1994, Sag et al. 2003), and started as a computational project through the LinGO initiative at CSLI, Stanford, using the LKB platform (Copestake 2002), which is a general platform with the format of typed feature-structures (TFS), and has integrated in it a format of semantic representation called Minimal Recursion Semantics (‘MRS’; cf. Copestake et al. 2005). Before year 2000 there were three grammars in this framework, the English Resource Grammar (ERG), the Japanese grammar Jacy, and a German grammar. Essential to the development of further grammars in the family was the ‘HPSG Grammar Matrix’ (‘the Matrix’; see Bender et al. 2002, 2010), which was mainly based on ERG, and had its first phase of deployment during the EU-project DeepThought (2002-4). The grammar family is currently developed within the frame of the DELPH-IN consortium, and will in the following be referred to as the ‘DELPH-IN grammars’.

The DG to be discussed was started in 2001, by linguists versed in Generative Grammar since the late 60ies, and formal semantics (‘Montague Grammar’) since the mid 70ies. From the mid 80ies the group developed a computational lexicon (under the acronym ‘TROLL’, see Hellan et al. 1989), mainly associated with research within ‘consolidated GB’. In the late 90ies the group reoriented itself towards HPSG, and started the DG as part of the LinGO initiative with the LKB platform. The DG was the first grammar to be built on the Matrix, during the EU-project DeepThought (2002-4), and despite never receiving very substantial funding, it has retained a place among the medium-large DELPH-IN grammars. We can distinguish four main phases in its development:

 Phase 1, the Grounding phase (2001-04), 
 Phase 2, the Semantic Expansion phase (2005-07), 
 Phase 3, the Cross-Linguistic Coding phase (2008-10), 
 Phase 4, the Interoperability phase (2010-14).

Phase 1 resided in the building of a basic core grammar around the Matrix skeleton (using the Matrix versions 0.1 – 0.6, as they developed; this included the MRS system). This stage included the accommodation of a 80,000 entries lexicon imported from the previously established resources TROLL and NorKompLex , where a verb valence code and a code for inflectional paradigms constituted major parts.

Phase 2 resided in the development of a fine-grained ontology and computing system of spatial and temporal relations, amenable to grammatical systems across languages and typologies, and a detailed semantics of comparative constructions. The grammar was also used as a part of a small Norwegian-Japanese MT system. In this period, the inflectional system was thoroughly revised. Main publications from this period are: Hellan and Beermann (2004), Bermann et al. (2004), Beermann and Hellan (2005).

Phase 3 was devoted to a thorough revision of the valence code, to accommodate a cross-linguistically defined classification system of valence and construction types. Main publications from this period are: Hellan (2008), Hellan and Dakubu (2009, 2010)

Phase 4 can be divided into the following themes:

A. Deploying the grammar in ‘external’ applications: a ‘Grammar Sparrer’, as described in Hellan et al. 2013, accessed at A Norwegian Grammar Sparrer. This is a construct along the lines of Bender et al. 2004, and Suppes et al. 2014, falling within the overall initiatives described in Heift and Schultze 2007, where specific types of grammatical mistakes are accommodated by ‘mal-rules’ in an extended ‘mal’-version of the grammar, and parses involving such mal-phenomena are reported to the user as tutoring instructions. This system has been running as a webdemo since 2011.

B. Exporting information from the grammar to independent resources:

1. A valence bank, which, with the same exporting strategy as for Norwegian, contains also two other languages, constituting the first instance of an in depth Multilingual Valence repository. In essence, the valence code used in verbal lexical types (cf. 3.2 below) is expanded to alternative and more easily inspectable formats, and the verb lexicons of the languages involved are imported into a database organized according to the newer codes, and searchable in terms of these codes. (See Hellan and Bruland 2013, and a web access at Multilingual Verb Valence Lexicon.)

2. A POS-tagger reflecting the lexical inventory of the grammar, useful for lexical acquisition from new text ( (http://regdili.idi.ntnu.no:8080/webtagger/tagger )).

3. A simple Reasoner over movement and spatial information exported from the MRS. (See Bruland 2013.)

C. Updating the soft- and hardware basis of the grammar (in particular with ACE ), and the construction of a server carrying web-facilities for the grammar itself, as well as the applications and resources just mentioned.

D. General regimentation of the grammar code.

E. Adapting the semantic system – MRS – according to the UTool restrictor on MRS objects, enabling the Reasoner mentioned above. (See Bruland 2013)

F. Aligned with the latter point, exploring ways of simplifying the grammar definition code, which in its Matrix shape is a remnant from a far more complex AVM notation system (from Pollard and Sag 1994) than is currently motivated. (See Hellan, forthcoming)

Point F is developed along a scheme of building small grammars for the exploration of specific features or alternative designs. A full DG as such is a very ‘stiff’ construct where no change can be made without repercussions elsewhere in the structure, hence no major change is recommended to be undertaken unless it has been explored in a smaller format.

Through points A and B, phase 4 addresses the third desideratum mentioned above, interoperability, and also the fourth desideratum, sustainability, in that the porting of information from the grammar into other applications is in effect probably the most efficient way of securing a maintained ‘life’ of the information. The grammar files as such will be stored, and their linguistic content and the computational environment will be documented, still it is through being used that information can be consistently improved. Points A and B are at the same time the ‘outreach’ aspects of the development, i.e., the respects in which the grammar makes itself useful. A given usage may well be the full motivation for the development of a grammar, although for DG grammars as ‘generic’ devices, their success in the usage domain should in principle be measured by how well they go into a multitude of applications; this is also the ambition behind the present grammar. The fact that such deployment of the grammar starts taking place seriously only after nearly 10 years of development, is to some extent a matter of coincidence, but attests to the complexity of such an object before it is ready for ‘deployment’.

More ‘immanently’ speaking, and returning to desideratum 1, the main duty of a DG is to accommodate in ‘analytic depth’ any sentence of the language for which it is defined, covering not only ‘core grammar’, but the whole array of constructs that can be used in texts of the language: ‘fragments’, abbreviations, interjections, and much more. With ‘in analytic depth’ one will normally include at least the following factors of morpho-syntactic structure and parameters of functional and semantic analysis (in addition comes a lexicon of significant size):

 - syntactic argument structure (or valence:  whether there is a subject, an object, 
 a  second/indirect object, etc., referred to as grammatical functions); 
 - semantic argument structure, that is, how many participants are present in the situation depicted,
 and possibly also which roles they play (such as ‘agent’, ‘patient’, etc.);
 - linkage between syntactic and semantic argument structure, i.e., which grammatical functions 
 express which roles; - identity relations, part-whole relations, etc., between arguments;
 - aspect and Aktionsart, that is, properties of the situation expressed by a sentence in terms of 
 whether it is dynamic/stative, continuous/instantaneous, completed/ongoing, etc..

If these, and other, factors can be addressed in a notional and formal system so cross-linguistically articulated that a grammar of a language L1, through its formal encoding, exposes a value or set of values relative to the above parameters within a matrix of corresponding values for languages L2, L3, …, etc., then one is on the track of attaining desideratum 2. That is, for every parameter chosen, each specification within the grammar of L1 for that parameter would project a value for L1 into a matrix where corresponding values for other languages are represented for that parameter. The field is not remotely close to reaching this stage, but it is again – like for desideratum 1 - an idea that one feels is reasonable. To a certain extent, the multilingual valence base mentioned under phase 4, point A, is an illustration of the point.

References

Beermann, D and L. Hellan. 2004. A treatment of directionals in two implemented HPSG grammars. In Stefan Müller (ed) Proceedings of the HPSG04 Conference, Katholieke Universiteit Leuven. CSLI Publications /http://csli-publications.stanford.edu/

Bender, E. M., D. Flickinger, and S. Oepen. 2002. The Grammar Matrix: An open-source starter kit for the rapid development of cross-linguistically consistent broad-coverage precision grammars. In Proceedings of the Workshop on Grammar Engineering and Evaluation, Coling 2002, Taipei.

Bender, E. M., D. Flickinger, S. Oepen and A. Walsh (2004). "Arboretum: Using a precision grammar for grammar checking in CALL," in Proceedings of the InSTIL/ICALL Symposium 2004, Venice, Italy.

Bender, E. M., Drellishak, S., Fokkens, A., Poulson, L. and Saleem, S. 2010. Grammar Customization. In Research on Language & Computation, Volume 8, Number 1, 23-72.

Bruland, T. 2013. Building World Event Representations From Linguistic Representations. PhD dissertation, NTNU.

Copestake, A. 2002. Implementing Typed Feature Structure Grammars. CSLI Publications, Stanford.

Copestake, A., D. Flickinger, I. Sag and C. Pollard. 2005. Minimal Recursion Semantics: an Introduction. Journal of Research on Language and Computation. 281-332.

Heift, T., and M. Schulze. (2007). Errors and Intelligence in Computer-Assisted Language Learning: Parsers and Pedagogues. Routledge, New York.

Hellan, L. 1988. Anaphora in Norwegian and the Theory of Grammar. Kluwer.

Hellan, L. and D. Beermann. 2005. Classification of Prepositional Senses for Deep Grammar Applications. In: Kordoni, V. and A. Villavicencio (eds.). Hellan, L. 2007. On 'Deep Evaluation' for Individual Computational Grammars and for Cross-Framework Comparison. In: T.H. King and E. M. Bender (eds) Proceedings of the GEAF 2007 Workshop. CSLI Studies in Computational Linguistics ONLINE. CSLI Publications.

Hellan, L.., L. Johnsen and A. Pitz. 1989. TROLL. Ms., NTNU

Hellan, L. (2008). Enumerating Verb Constructions Cross-linguistically. In Proceedings from COLING Workshop on Grammar Engineering Across frameworks. Manchester. http://www.aclweb.org/ anthology-new/W/W08/#1700.

Hellan, L. and M.E.K. Dakubu (2009): A methodology for enhancing argument structure specification. In Proceedings from the 4th Language Technology Conference (LTC 2009), Poznan.

Hellan, L. and M. E. K. Dakubu, 2010: Identifying Verb Constructions Cross-Linguistically. Studies in the Languages of the Volta Basin 6.3. Legon: Linguistics Dept., University of Ghana

Hellan, L. and T. Bruland 2013. Constructing a Multilingual Database of Verb Valence. NoDaLiDa, Oslo, 2013.

Hellan, L. T. Bruland, M. Sandøy, E. Aamot. 2013. A Grammar Sparrer for Norwegian. NoDaLiDa, Oslo, 2013.

Pollard, C. and Sag, I. (1994). Head-Driven Phrase Structure Grammar. Chicago University Press.

Suppes, P, T. Liang, E.E. Macken and D. Flickinger (2014) “Positive technological and negative pre-test-score effects in a four-year assessment of low socioeconomic status K-8 student learning in computer-based Math and Language Arts courses ", Computers & Education, 71, pp. 23-32 (http://www.sciencedirect.com/science/article/pii/S0360131513002637).

Test suites

Complete test suites for basic verbal constructions are found in

• Verbconstructions Norwegian - intransitive,
• Verbconstructions Norwegian - transitive,
• Verbconstructions Norwegian - ditransitive and copular.