keep-with-next.within-page="always"in section 1.1 keep-with-next.within-page="always"in section 1.1contact: andy_black@sil.orgPROBLEM1PROBLEMSapir (1921) has said that “All grammars leak” meaning that as much as we want to believe that all published grammatical descriptions are watertight, there is always data that the linguist has not seen or has chosen to ignore (residue). The description may adequately explain what is happening in a carefully selected set of data or the description may adequately account for the majority of the data. But still there is residue that is not reported. How can the reviewer of a grammatical description ever obtain access to the residue, or ever be certain that the description would actually work for a totally new text gathered from a native speaker, or even that it would even work for the whole data corpus the linguist was studying? A new kind of linguistic analysis and description is needed.Morphology1.1 MorphologyThe real test of a description of the morphology of a language is whether it can be used to parse any word in the language appearing in a natural text while at the same time rejecting ungrammatical wordforms. The ultimate goal is that the description could be empirically verified by implementing it in an automated parser that would correctly parse any wordform occurring in natural text. FLEx enables such synchronization between description and automatic parsing as it uses the same underlying data to generate both a morphology sketch and an automated parser for testing the accuracy of the analysis on natural texts.The goal of morphological research is not only to unambiguously parse all the words of a language into their constituent morphemes, but also to account for the whole structure of the wordform. Ideally, Philippine language features like voice/focus and aspect could be linked to the morphemes that bear them and then percolate up through the parsing process to become a part of the description of the whole wordform. This information can then be used by syntactic parsers that would operate on the analyzed wordforms. This is especially important for Philippine languages where voice affixes relate to the semantic roles of the nouns in the sentence. A syntactic parser that will work on the output of the morphological parser will need this feature information also, not just the final syntactic category of the wordform.FLEx is the only software package that integrates a morphological parser, lexicon, and analysis tools with the ability to automatically generate structured morphology sketches. Unfortunately, even though FLEx has been available for a few years, uptake has been slow among Philippine linguists. Of those that are using it for dictionary purposes, none have made a serious attempt at a computational description of the morphology. Because of the degree of similarity between affixation systems of Philippine languages, having an empirically tested morphology model for one Philippine language should prove helpful for other Philippine language researchers and thus simplify their task.FLEx automatically generates a sketch of the morphology based on the information in the database. This tool has the potential for aiding both linguists and linguistic consultants in identifying problems and inconsistencies in analysis as well as being a basis for a publishable morphological description. To date, however, the parsing tool has not yet been used for checking analyses and the morphology sketch has not yet been used for describing a Philippine language. Research identifying the strengths and weaknesses of using this method of analysis and description can help clarify its usefulness for other Austronesian languages as well.Syntax1.2 SyntaxThe real test of a syntax description is whether it can account for the syntax of any grammatically correct sentence in that language. When prose-only descriptions of the syntax of a language are written, it is hard to ascertain what percentage of the language phenomena is being described. If, however, the syntax description is formalized with a set of formal PSRs (Phrase Structure Rules) and natural language texts are used for testing the rules, it may be possible to state what portion of the syntax is accounted for by the description.The team at the Center for Language Technologies led by Dr. Rachel Roxas of De La Salle University in Manila is seeking to develop computational tools and ultimately machine translation systems to and from Tagalog. Members of this team have mentioned “the minimal work done on the development of a computational grammar for the Filipino language for the development of robust and industrial strength natural language analysis and technologies” Alcantara and Borra (2008a). Many approaches have been proposed and prototypes have been developed for automatic generation of PSRs from various text collections, but as of this time, all of the final results have been labelled “insubstantial” by Borra.11Personal communcation (Nov. 11, 2008).The lack of a complete set of PSRs for Tagalog is a significant hindrance to NLP research in the Philippines. The same problem exists for other Philippine languages, especially the minority languages. But because of the similarities of syntax of many Philippine languages, it is hypothesized that PSRs modelling general language characteristics for any Philippine language could contribute insights into the development of PSR grammars for other Philippine languages. Thus a set of basic PSRs for any Philippine language could be used as a starting point for developing PSR grammars for other Philippine languages. However, as of this time, no such published grammar exists in a current linguistic formalism.The PAWS Starter Kit could be a significant step forward toward this goal of developing Context Free PSRs for Philippine languages. If the developer’s claim is really true that 75% of the syntax of any language can be modelled in PSRs generated by just answering the questions in the PAWS Starter Kit, then this would be a significant step forward for all those doing research in this area. The problem is that no one has attempted to use the system for any Philippine language. Determining the percentage of phenomena handled by the automatically generated rules and performing an analysis of what went wrong and right in the syntax parsing process would be most helpful not only to the developers of PAWS but also to Philippine linguists considering whether to use it.The syntax sketch that PAWS generates would seem to be a helpful tool for linguistic description, but it has not been used for a syntactic description of any Philippine language. This thesis provides an initial syntactic description of a Philippine language using the PAWS generated sketch. Discovering the areas where the linguist needs to adjust the description will be the most helpful in evaluating the usefulness of this tool.ReferencesReferencesAlcantara, Danniel, and Allan Borra. 2008a. “Constituent structure for Filipino: Induction through probabilistic approaches.” Proceedings of the 22nd Pacific Asia Conference on Language, Information and Computation (PACLIC22). Cebu, Philippines.Sapir, Edward. 1921. Language: an introduction to the study of speech. New York: Harcourt Brace & Company.