This item has the following downloads:

aymaraNSF2007v2Narrative ( PDF )

    Overview
        Page 1
    The project investigators and other key personnel
        Page 2
    The languages
        Page 3
        Page 4
    Source data
        Page 5 (MULTIPLE)
    The accessible linguistic database
        Page 6
        Page 7
        Page 8
        Page 9
    Web service architecture, user interface and interoperability
        Page 10
        Page 11
    Work plan: Roles and responsibilities
        Page 12
    Project timeline and procedural plan
        Page 13
    Broader impact and intellectural merit
        Page 14
    Planned future projects
        Page 15

An Accessible Research Database for the Endangered Jaqaru and Kawki Languages
M.J. Hardman (P.I.), Elizabeth Lowe McCoy(Co-P.I.)
Howard Beck (Co-P.I.) with Sue Legg, Techical Liaison and Evaluation Consultant
The University of Florida, Center for Latin American Studies


Overview

This project will document the endangered Jaqaru and Kawki languages, which are members of the Jaqi
family of languages (Jaqaru, Kawki and Aymara). Jaqaru is spoken in the Andes Mountains of Peru by a
few thousand people resident in Tupe, Yauyos, Lima, Peru and in the cities of Lima, Huancayo, Chincha
and Caiete. Diaspora Jaqaru speakers are located in small numbers throughout Latin America, the
United States and Europe. The District of Tupe consists of three interrelated communities: Txupi (Tupe),
Aysha (Ayza) and Qullqa (Colca). The population of the district is very low, having declined in the last
twenty years due to the migration of young people to the cities and the increase in terrorism. The
highland plaza is 9,000 feet above sea level and access to the area is by foot and pack animal. Kawki is
a dying language spoken by only a few people around Cachuy, Chavin, and Lima, Peru. Access to high
quality research materials for Jaqaru and Kawki is viewed as a priority by the people of Tupe. In July
2007 a town meeting was held with the Regional government and the Peruvian Ministry of Education to
protest the scarcity and poor quality of educational materials in the languages. The meeting resulted in a
resolution of support to improve bilingual education. The earthquake of August 2007 destroyed Tupe and
caused damage to roads and adjacent villages, placing the languages in further danger of disappearance.
This project will build on the life work of Principal Investigator, M.J. Hardman, who first began her work on
the Jaqaru, Kawki and Aymara languages in 1958. Her first Jaqaru Grammar was published in 1966 by
Mouton; the Spanish version was published in 1983 by the Instituto de Estudios Peruanos; the updated
Hardman Jaqaru grammar was published in 2000 by Lincom Europa. She is the author of extensive
publications on the Jaqi languages, including Aymara: Compendio de Estructura Fonol6qica y Gramatical
(ILCA, 1988). The project will build on Dr. Hardman's past fieldwork and well as new fieldwork in Peru, to
transform a corpus of 50 notebooks of texts, corresponding audiotapes, and related linguistic data into an
accessible linguistic research database of the Jaqaru and Kawki languages. A unique feature of the
database will be a dictionary that will be produced with interactive collaboration from the Jaqaru and
Kawki speaking community, enabled by a Wiki function. The work responds to the call for interoperability
of tools and common data identification standards made by Gary Simons at the Summer 2007 TILR
workshop at Stanford University. It builds on existing computational and linguistic work, and conforms to
the recommendations and standards proposed by the E-MELD consortium. The Jaqaru project continues
the work currently being concluded on a linguistic database developed with a grant (M.J. Hardman, P.I.)
from the U.S. Department of Education International Education and Graduate Programs (Title VI) titled:
"The Aymara E-Learning Project: Using the Internet to Preserve and Promote an Indigenous Language."
database envisioned for this project will have many of the features of the Aymara database, however
unlike the Aymara project, which has instructional objectives, the Jaqaru-Kawki database will be designed
for the use of researchers and native speakers to document and preserve these endangered languages.
It will be multilingual (Jaqaru, Kawki, English and Spanish). It will contain dictionaries of the two
languages, audio, digital photographs, and the grammatical and linguistic elements of the languages. The
work to index and digitize the original materials, to develop the database and to translate the linguistic
work into English and Spanish will be conducted primarily at the University of Florida, in Gainesville,
Florida. A coordinator in Peru will be responsible for the collection of additional materials, the transcription
of the audio tapes, as well as the deployment and training of local counterparts in the project
implementation, the development of the interactive dictionary and the use of the database. The broader
impact of this project will be to preserve and make available in four languages the grammar and
dictionaries of two endangered Andean languages for linguistic research and for the use of heritage
speakers in a strategically important world region. The intellectual merit of the project resides in the
multifunctional nature of the database, which has features that are distinct from the SIL LinguaLinks
project and the FLEx project, or work done by Charniak and Collins. The project will contribute to the
multi-university community work on the standards to ensure the robustness, accessibility and
interoperability of electronic archives for endangered languages. The Jaqaru-Kawki resources for the
project will be catalogued on appropriate metadata forms and deposited in the Archive of Indigenous










Languages of Latin America at Texas, Austin (AILLA) and the accessible linguistic database will be
mapped to the University of Florida Digital Collections with appropriate metadata and ontology standards
for digital archiving and will become part of a forthcoming Aymara Collection within a larger South
American collections library at the University of Florida.
The project description will be organized as follows. After introducing the project investigators
and key personnel in section 1, we present some background information on Jaqaru and Kawki is
presented. Section 2.1 describes the ethnographic setting; 2.2 discusses the endangered status of the
languages; 2.3 surveys existing documentation for the Jaqi and Kawki languages. Section 3 describes
the linguistic resources that will be digitized and analyzed for the primary product of this project: the
accessible linguistic database for the Jaqaru and kawki languages. Section 4 describes the archiving plan
(how the material maps to AILLA and to the University of Florida Digital Collections) and samples of the
metadata. Section 5 describes the accessible linguistic database, tools and access standards and issues
and compares and contrasts the proposed work to work done elsewhere. Section 6 discusses the Web
Service Architecture, User Interface and Interoperability. The responsibilities of the project collaborators
and the project timeline are specified in section 7. Section 8 provides a summary of broader impact,
indigenous use and intellectual merit of the project. Section 9 briefly outlines planned future projects,
including a plan for testing in other contexts.

1. The Project Investigators and other key personnel
This project is an interdisciplinary, international collaboration that brings together several areas of
research interest and expertise at the University of Florida and with local institutions in Peru. It also
draws upon the life work of the leading scholar on Jaqi languages working in the field of anthropological
linguistics.
M.J. Hardman (P.I.) is a professor of Linguistic Anthropology at the University of Florida and is
an affiliate faculty member of the Center for Latin American Studies. She is one of the pioneers of
research with the Jaqi languages from an anthropological linguistic viewpoint, a lifelong work that she
began in the 1950's as a field researcher in Peru. She has extensive publications on Aymara, Jaqaru and
Kawki, most recently her Jaqaru Grammar (Lincom Europa, 2000). She is credited with being the first
outside researcher to discover the existence of Kawki and the relationship with Jaqaru and doing the first
linguistic research on them, in collaboration with her husband, Dr. Dimas Bautista Iturrizaga, D.V.M., who
in the 1940's sought a way to write his language (Bautista, 2000). In 1965, with Dr. Julia Elena Fortun,
she founded the Instituto Nacional de Estudios Linguisticos (INEL) in La Paz, Bolivia, as a dependency of
the Directorate of Anthropology of the Bolivian Ministry of Education and Culture. As a Fulbright-Hays
Professor, she taught courses to prepare human resources in linguistics in Bolivia. She was the Director
of the Aymara Language Program at the University of Florida from 1969-1990, which was funded by a
grant from the U.S. Department of Education Title VI program. Dr. Hardman continues to do research on
the Jaqi languages and spends several months each year in the region. Dr. Hardman is P.I. of the U.S.
Department of Education Title VI Grant, "The Aymara E-Learning Project (2004-2007)." Her current
research also involves language and gender and the patterning of worldview in language.
Elizabeth Lowe McCoy (Co-P.I.) is Associate Director for Program Development and Distance
Learning at the U.F. Center for Latin American Studies and founding director of the Partnership in Global
Learning, an international distance learning initiative funded originally by the Lucent Foundation and Bell
Labs. She is establishing a Research and Training Program and Academic Specialization in Indigenous
Languages and Language Policies of the Americas within the Center. Dr. Lowe is the Co-P.I. of the U.S.
Department of Education Grant "The Aymara E-Learning Project," and is responsible for overall fiscal
and administrative management of the project, serving as leader of the interdisciplinary team. Former
Executive Director of the U.F. International Center (1991-1998), and founder of the U.F. certificate
program in translation studies, she has been a project manager for large and complex international
programs and is an expert on the use of technology for foreign language and translation instruction. Her
publications are in the field of Latin American literature, language and culture.
Howard Beck (co-P.I.) is a professor in the Agricultural and Biological Engineering department at
University of Florida, where he has been on the faculty since 1990. He has been working at the UF
College of Agriculture since 1977 and has involved in a broad range of information technology-related
projects. He completed his Ph.D. in 1990 in the UF Computer and Information Sciences department in
1990, within the Database Systems Research Center. Dr. Beck's research combines artificial intelligence
techniques with database management in order to create knowledge management systems that can










organize vast collections of knowledge within a particular domain. He pioneered the use of object
database management systems to develop digital libraries in agriculture and natural resources. More
recently the work has evolved to incorporate formal ontologies as the core of a database management
system, resulting in his development of the Lyra ontology management system. A unique feature of Lyra
is its support for natural language processing, and current work includes development of a database on
the Aymara language using Lyra. Other successfully database systems developed and deployed by Dr.
Beck include EDIS (Extension Digitial Information System, which is a digital library of Extension
publications for the College of Agriculture, FAWN (Florida Automated Weather Network), DISC (Decision
Information Systems for Citrus), SPDN (Southern Plant Diagnostics Network), and CBC (Crop Bioscurity
Curriculum). These applications incorporate a variety of techniques including expert systems, computer
simulation, eLearning, and information retrieval.
Sue Legg (Technical Liaison and Evaluation Consultant) was the Associate Director of the
Office of Instructional Resources (OIR), now the Office of Academic Technology at the University of
Florida from 1980-1995 and Director from 1995-2001. In 2002, she also became Director of the Center for
Instructional and Research Computing Activities (CIRCA) and U.F. Coordinator for Distance Education. In
2002, Dr. Legg retired from her administrative responsibilities and joined the Center for Latin American
Studies, where she was Director of the Partnership in Global Learning (PGL) until 2004, when she
assumed the role of Research Director. PGL is an international university and K-12 consortium whose
mission is to develop and implement online learning technology and training. Also in 2004, she assumed
responsibility for technical coordination of database development for the "Aymara e-learning project." Her
area of specialization is in research methods, measurement, and evaluation. She was principal
investigator for a number of contracts and grants from the Florida Department of Education and national
foundations. Her publications are in the area of assessment and evaluation. She also continues to serve
as the assessment consultant for the Florida Bar Board of Legal Certification and Education.
Yolanda Nieves Payano Iturrizaga (Peru Coordinator) is a linguist who has been appointed by
the Regional Government of Lima to supervise bilingual education in Tupe, Peru at the Instituto Superior
Pedag6gico Yauyos in the Department of Lima. Ms. Payano studied linguistics and languages at the
Universidad Mayor de San Andr6s (USMA) in Bolivia and studied English in the United States for one
year, while assisting with a Field Methods course at the University of Florida. Dr. Dimas Bautista
Iturrizaga, D.V.M (Jaqaru Language Resource Consultant) was mayor of Tupe, Province of Yauyos,
Peru from 1984-1986 and has been involved with efforts to preserve the Jaqaru language since the
1950's. He met the famous linguist Kenneth Pike when he approached the Peruvian Ministry of
Education, in search of someone who could help him write his language. Trained as a doctor of veterinary
medicine, Dr. Bautista began to publish on the Jaqaru and Aymara languages as a result of his work in
Puno with the alpaca, for which he discovered and developed a vaccine that made possible herds of
alpacas and thus today's alpaca wool industry. While inoculating the animals, he noticed someone had
used an Aymara term to refer to pasturing (awata in Jaqaru, awatina in Aymara) which he understood and
replied in Jaqaru This resulted in his publication "Tupe y el Jaqaro o Kawke" (1959). Currently in
preparation is his book on the history of Tupe, relating what he was told by his elders ( Mark Qillqa: Tupe,
Resefa hist6rico-cultural del pueblo de Marka afo 750-2007," Instituto de Estudios Peruanos, 2007). Dr.
Bautista has been a principal consultant to Dr. Hardman throughout the years and frequently the source
of her publications.

2. The Languages

2.1 The Ethnographic Setting
The Yauyos valley is a very steep, rugged valley that has long impeded the intrusions of outsiders because it
is so difficult to negotiate. It is also the most linguistically varied area of the Andes (cf., Alfredo Torero). It is
thought that the Incas never entered the valley, and even the Spanish were slow to do so. Eventually the Spanish
penetrated the area, and in 1761 the Virrey Amat issued an establishment document for Tupe. The Jaqi people
are primarily farmers, herders, weavers, musicians, dancers and marketers. The Jaqi civilization (400-1000 C.E.)
established both a complex system of irrigation by canals and a complex system of mercantile exchange,
involving the construction of roads. Some of the market relationships still function today and the road system is
clearly visible in ruins. One cultural invention was the archipelago land holding system where an individual would
attempt to hold plots of land, however small, in as many different ecological niches as possible. The land is steep
so that very different crops can be cultivated not far apart by going up or down the mountain. Cultivating crops at










different elevations ensures diversity in what they produce and protects against weather and pests. The
archipelago system remains the desired form of land holding. This system, together with extensive marketing,
today specializing in cheese, means a great deal of movement as people go about ordinary tasks. Travel to
Watxuqu (Catahuasi), some 15 miles distant and 5000 feet lower, and returning the same day, is routine. School
has always been attractive to Jaqi people and there are today many professionals, including teachers, doctors,
veterinarians, engineers and others who have come from Tupe. The modern city has become one more
ecological niche within the Jaqi archipelago system of land holding, with foodstuffs and weavings coming from
Tupe and manufactured goods, especially electronics such as radios and batteries returning from the city.
Jaqaru is a member of the Jaqi family of languages which also includes Kawki and Aymara. Jaqaru is spoken
by a few thousand people around Tupe, Yauyos, Lima, Peru and in the cities of Lima, Huyancayo, Chincha and
Caiete, with a few diaspora speakers scattered throughout Latin America and in the United States and Europe.
The district of Tupe consists of three interrelated communities; Txupi (Tupe), Aysha (Ayza) and Qullqa (Colca).
The current population in the District is very low, consisting of only around 600 people. It has dropped from 5000
in the past 100 years, due to the outward migration of young people and the advent of terrorism. Kawki is a dying
language spoken by only a very few people in and around Cachuy, Yauyos, Lima, Peru. Aymara is spoken by
two to three million people, the first language of a third of the population of Bolivia and the major native language
in Southern Peru and northern Chile.

2.2 Endangered Status
When Hardman first began her study of the Jaqaru language in 1958 there were still monolingual speakers of
Jaqaru and quite a few people who had learned Spanish only late in life, whose knowledge of Spanish was
limited. Today all the young people of Tupe are bilingual and a number of children now do not speak (but still
understand) Jaqaru. There are no living monolingual speakers; even the oldest living bilinguals are fully fluent in
both languages. Jaqaru, therefore, is an endangered language. In the early 1900's, the elders of Tupe sought to
establish a high quality elementary school. Over a period of about 30 years, they succeeded. They brought
books for the school from Boston, materials for a physics and chemistry lab, a globe and recruited teachers from
the outside. Tupe became the educational center of all of south Yauyos; students came as boarders to study in
Tupe. In an era when the eradication of native languages was the goal, children in the school were punished for
speaking Jaqaru. Within two generations, by the middle of the century, virtually all speakers were at least
bilingual, though there remained some for whom Jaqaru was indeed dominant, who spoke a rudimentary Spanish.
That is no longer the case. During the 1980's, Tupe suffered from the intrusion of Shining Path terrorists. A
massacre led to migration, which accelerated the shift from Jaqaru to Spanish. In the late 1990's, when Shining
Path was no longer a threat, children no longer spoke Jaqaru as a primary language. Those who were children at
the beginning of the terrorist assault, including some current school teachers, now wish to recover the language,
even though their own control of the language is sometimes shaky. Some people, including the high school
students in Tupe itself, are hoping that bilingual education will preserve the language as part of their endangered
cultural heritage. Recent regional government initiatives seek to strengthen bilingual education. In July 2007 a
town meeting was held with the Regional Government and the Peruvian Ministry of Education to protest the
scarcity and poor quality of educational materials in the language. The meeting resulted in a resolution of support
("Acta") to improve bilingual education and encourage the language training of heritage speakers. The
endangered status is even more grave due to the August 2007 earthquake that destroyed Tupe and caused
damage to roads and adjacent villages.
The extinction of the Kawki language was provoked by the discovery of a miraculous image, el Senor de
Cachuy. The religious icon attracted hordes of pilgrims, and unleashed a number of changes that threaten their
language and culture In May of every year, the town of 200 swells to over 20,000 pilgrims, including evangelical
missionaries whose actions hasten the shift from Kawki to Spanish. By the middle of the last century, the
language was spoken only by the elderly, with the exception of one individual who had been raised by his
grandmother, Valerio Luciani Ascencio, on whose shoulders the future of the language now rests. He is almost
60, but he has been, for the last 30 years, attempted to teach Kawki to school children, laboring in the absence of
official recognition and financial support.


2.3 Previous Documentation
The first person to be interested in a study of the language was Dr. Dimas Bautista Iturrizaga, D.V.M., who in
the 1940's sought a way to write his language. The story is told in his forthcoming book on the history of Tupe
(Bautista, 2007). In recent years two young women have begun to work with the language, the linguist Yolanda










Nieves Payano Iturrizaga (Payano, 1988) and Neli Belleza Castro (Belleza, 1995). There is also a periodical that
publishes some material in Jaqaru (Ramirez, 1989-to present). Also, Dr. Jose Matos Mar collected some
materials in the 1950's. The meager results of early studies, are referenced in Hardman's 1966 and 1983 Jaqaru
grammars. She has written numerous articles on Jaqaru and Kawki, referenced in the bibliography. Over the
years Hardman has continued to work with Jaqaru as well as with the sister languages, Aymara and Kawki.
Included in the Kawki documentation are bilingual Kawki-Spanish primers (Hardman, 1982,1983), Kawki texts for
school use (Hardman, 1981, 1980), a Kawki-Castellano-English trilingual dictionary (Hardman, 1986) and an
alphabet of the Jaqaru, Kawki and Aymara languages (Hardman, 1991).
Hardman's recent Jaqaru Grammar (2000) is the most systematic attempt to document the language.
The grammar illustrates some of the language's unique features and provides the organizational framework for
the proposed accessible research database, to be adapted from the current Aymara database project (U.S.
Department of Education Title VI, 2004-2007). The phonemic system distinguishes 36 consonants but only 3
vowels. Vowel dropping is significant, complex and pervasive, marking case and phrase structure as well as
style. The language makes extensive use of morphology, with all verbs carrying several suffixes. Syntax is
morphologically marked; verbal person suffixes mark simultaneously object/subject; data source is marked at all
levels of grammar. Within the nominal system inclusive/exclusive and humanness are marked.

3. Source Data
The primary source of the data for the project will be the Jaqaru field materials collected by Hardman during her
fifty years of research in the region, along with new text, and audio and photographic materials that will be
collected and digitized by the coordinator in Peru. The materials include:
1. 50 hand written field notebooks containing transcriptions of existing audiotapes, notes about the content of
the tapes, grammatical analysis and illustrative hand drawings.
2. corresponding audiotapes of native speakers of Jaqaru and Kawki. Many of these recordings are of now
deceased village elders.
3. Over 1000 slides taken between 1959-1975 of Jaqaru and Kawki speakers, villages, farms, market
scenes, homes, and schools.
4. Interactive dictionaries of the two languages, allowing for user collaboration with a Wiki function.

Figure 1 shows a screen shot of one of the field notebooks.

4. The Archiving Plan

4.1 Archive of Indigenous Languages of Latin America at the University of Texas, Austin (AILLA)

A sample of the metadata form is in the Appendix


4.2 University of Florida Digital Collections
Source Data will be digitized by the University of Florida's Digital Library Center. Text page images will be
scanned at 300 dpi, while graphical images will be scanned at 600 dpi. Resulting images will be mastered in the
TIFF v.6 uncompressed format, with derivatives for Internet use in JPEG and zoomable JPEG 2000 formats.
Searchable text, generated by double-key method to 100% accuracy, will be procured from the Center's vendor.
Audiotapes will be sampled at 24-bit 96.0 kHz and retained as WAV uncompressed format, with derivatives for
Internet use in down-sampled 16-bit 44.1 kHz MP3 format. The high-resolution WAV will also be available for
download. Internet accessible versions of all digitized source data will be available in the University of Florida
Digital Collections' Jaqi Collection (http://www.uflib.ufl.edu/ufdc/?s=aymara). The Digital Collections (UFDC)
utilize both Greenstone digital library and Aware JPEG 2000 data stores, as well as a locally designed audiovisual
store. All files and metadata will be fully searchable and available freely through the Internet. All digitized
resources will be supplied to the Project PIs for use in the Language Editor and other project applications.
The Center digitally archives all content including the projects executable files and data-tables, as
well as digitized resources with the Florida Digital Archive (FDA). FDA
(http://www.fcla.edu/diqitalArchive/) is a service of the Florida Center for Library Automation (FCLA),
based on the framework for Open Archival Information Systems (OAIS). For more information on FDA,
see http://www.fcla.edu/diqitalArchive/pdfs/IJDL article.pdf. Procedures for data validation, format
migration and redundancy are key; archived resources are validated against DTDs and checksums and










retained by FCLA's FDA, backed up on the Northwest Regional Computing Center in Tallahassee,
Florida, and copied to Data Center at Stanford University. FDA is funded by the State of Florida through
continuing allocations to FCLA. The service is free to the University of Florida.

5. The Accessible Linguistic Database

5.1 An Ontology Management System for Language Archiving. Central to the goal of building a
collaborative environment enabling team members to work together is an environment consisting of on-
line authoring tools that team members can use to create and edit data objects, and a core database
facility for storing these objects.

We are utilizing an ontology management system (OMS) to provide this environment, including facilities
for representing and archiving knowledge associated with the language and culture of Aymara. The OMS
provides a framework for integrating everything from raw data elements (sound recordings, transcripts,
images, video), to more abstract linguistic elements (morphemes, words, phrases, phrase patterns
(grammars), and dialogues). Logical relationships between any two elements are expressed using
ontology property relationships (i.e. how the data elements are related). While we have built a database
for Aymara, the system can be applied to any language.

The Lyra Ontology Management System. While ontologies are traditionally used to represent semantics
of words, we expand the use of ontologies to the level of full-blown database management system, that is
a database system that uses a formal ontology language (such as OWL, the Web Ontology Language [])
as the data definition language, rather than tables as used in conventional relational databases, or
general purpose persistent objects as used in object database management systems. The ontology
language provides a more natural and logical way of describing concepts and relationships than previous
database languages. Concepts are arranged taxonomically, through part/whole relationships, and
through other relationships such as needed to capture the connection, for example, between a phrase
and the original transcripts. Every abstract element is thus not only tied directly to original data, but such
abstractions can be generated based on generalizations obtained over raw and intermediary structures.
Thus the OMS provides a complete environment for modeling and storing all data objects and
relationships needed to build a linguistic database, including raw field data and cultural artifacts.

We have constructed an ontology management system called Lyra, that provides a number of basic data
management services. Lyra uses an ontology language for data modeling and representation. Lyra is
similar to OWL, and we have developed mappings between Lyra and OWL. Lyra also supports physical
storage management optimized to storage and retrieval of large numbers of objects. Lyra includes a
number of authoring tools for viewing, designing, and editing objects. The authoring tools provide a data
visualization environment, and tools can be customized to a particular application in order to provide
experts with tools that they can be most comfortable with. We are currently incorporating reasoning
capabilities for Lyra that support traditional ontology reasoning subsumptionn and classification), as well
as parsers and inductive learning algorithms (for use in grammar induction). The ontology reasoners will
thus provide a toolkit for searching the database, and discovering new categories and relationships by
analysis of new and existing data objects. Finally, the OMS publishes data using web services and XML
interfaces provide interoperability between internal Lyra data object structures sharable and sharable
XML data formats.

There are several advantages of using an OMS rather than a traditional relational database, or no
database at all (XML file system). First would be the more natural way in which an OMS models linguistic
(and other) data. The ontology language provides an way to model taxonomic relationships, and other
relationships among objects, with direct pointers among abstract generalizations and concrete data (such
as field observations). The object structure can model complexities of language elements. Though this
can be done with relational databases, it is much more difficult as these complex structures must then go
through an additional mapping to normalized relational tables in which the object structure and
relationships is no longer explicit. This can also lead to less efficient data retrieval []. Another advantage
would be the focus on reasoning, and in particular the computational complexity of reasoning processes
that operate on the data. It is known that the details of the data modeling language have a direct impact










on computational complexity of reasoners. The ontology language can be designed to capture just
enough detail, but not too much to lead to computational complexity problems (for example, OWL-DL, the
description logic version of OWL, is designed to balance this tradeoff). Finally, an OMS offers a variety of
services necessary to support security, data integrity, transaction management, and query processing.
XML files are ideal for exchange of data among different systems, but the database is needed to provide
operational functionality. One could also write custom tools that read and write XML files directly, but
they would need to duplicate the functions provided by an OMS.

Lyra was not developed solely for the Aymara project, but has evolved over nearly two decades of
research and development on combining databases with knowledge representation techniques from
artificial intelligence to build a system capable of organizing and storing knowledge in broad domains.
Lyra has been used successfully to model applications in agriculture [], environmental sciences [],
mathematics [], decision support systems [], and linguistics []. The ability to support natural language in
an integral fashion is one of the core advantages of this approach.

5.2 Tools. Lyra includes several authoring tools that provide data visualization environments for
browsing, creating, and modifying data objects. The tools are designed for use directly by subject matter
experts, and are used by members of our team to construct the database. Such tools can be developed
by anyone to access the database using web services (See next section). The tools are graphic
(providing visualization) and Web-based (run inside Web browsers using plugins), thus they form the
basis for a collaborative environment for creating data objects stored in and retrieved from a common
database. The tools can also be customized to meet the needs of experts working in a particular domain.

5.2.1 Language Editor

Figure (3) is a view of the LanguageEditor, a tool for browsing and editing the language database.
LanguageEditor is a Java applet that runs in any web browser that has the Java plugin. It connects
remotely to access the database over the Internet. Shown here is a complete phrase analysis including a
morphological analysis. The gloss for the dictionary entry for one of the suffixes is displayed at the bottom
of the screen.


The LanguageEditor is the primary tool used for constructing the Aymara database. It includes a section
for entering each unit in the training database (as shown in Figure (1). It also includes sections for phrase
patterns (grammar), categories (maintenance of grammatical categories), access to all phrases in the
database, and dictionaries for individual words and morphemes. Shown in the figure is an analysis of one
of the phrases appearing in a dialogue. Multilingual glosses are supported (in this application English and
Spanish). Facilities for recording and storing sound recordings of the phrases are shown, as well as an
image associated with this phrase. The phrase analysis includes a morphological breakdown, and each
word and morpheme couples directly to the dictionary. Note the markers and features associated with a
particular morpheme as shown in figure.












-_ilJ 1j


Utilities
PROJECT PHRASE PATTERNSI CATEGORIES PHRASES WORDS MORPEMES EDI
-I Dialog (Phrases)
L-oabel Enter Phrase: j Tata maya\yjiskt'asima [jsk\t'a\sl\ ma]
Introduction
Instructions English Gloss: IW. Sir, may I ask you a question?
Notes English Morph Annotation: O
Patterns
.Winus tyas, mama. Spanish Gloss: IW. Senor, disculpe una pregunta.
Spanish Morph Annotation: I2
J. Kunach mam?
W. Kuriiruran kunas sut Record Listen Load Sound File I Nts F l
J. Piruti Apas satawa. Ji
W. Jisa juparuwa, alujt
J. A.. alujamintux k"a is
W. ayattata' W tata mayay jiskt'asTma
3. anlw aya tl... k"a is
W. A... isa.
UJW. Nayax ukaks maya y jisk t'a si
J. Janit jutinkta?
W. Janiwa, ]ich"aklpunr I ma
J. Kawkinkritas mama? l>2p future
W. Istarusunirusanki'hi
, Nayax akankritwa ,,. Base Morpheme: ma root r suffix Ivy ]
W. Walikiw, uk'amax q'
C nominal C verbal stem derivation C verbal person derivation ( verbal inflectional C thematic C sentence C independent
English Gloss: 1>2p future
Spanish Gloss: I1>2p future


Move Up
Move Down
New
Delete Save


Concordance Features
Feature Value
source person 1
tense Future
:arget person 2

J


Figure 3. The LanguageEditor is a Web-based tool for creating the Aymara database. It is used on-line
by field workers.


5.2.2 ObjectEditor


The ObjectEditor is another authoring tool in Lyra, and illustrates how data are entered into the database
as conceptual units and examples. The ObjectEditor is a more general purpose object creation tool, in
contrast to the LanguageEditor which is customized for creating linguistic objects. We use ObjectEditor to
develop documentation for the Aymara grammar. Authors can sequence the same content in different
ways as well as add to or eliminate content in order to adapt it for different audiences. In our application,
the objects are organized by the issues being discussed. Alternative organizations of the same objects
lead to different perspectives. For example, it is also possible to categorize the content by grammatical
elements. From this approach, it would be possible to generate an index or several indices to highlight
research and/or learning issues.


Figure (4) shows both a presentational view and a conceptual taxonomic view of a set of objects
describing the introductory grammar. On the left is a column of topics (origin, nouns and verbs, etc).
arranged sequentially (arrows with plain heads represent sequence) within a grammar lesson for Unit I
Exercise Set I. This sequence is the order in which concepts are presented to a student in a linear
fashion. Labeled associations from the lesson (represented by the node, "I exercise set I lesson") to each
topic indicate the role of the object within the lesson (introduction, definition, conclusion). An expansion of
one of the concepts ("yes/no interrogative -ti") shows a text description of the concept written using a
multilingual text editor. The right side of the diagram shows an alternative view in which objects are
categorized within a taxonomy of grammatical terms (word, morpheme, sentence, etc.). Only the portion
of this taxonomy relevant to this particular set of objects is shown, but in fact the complete database is
much larger than this, grouping objects across all lessons, and integrating them with the complete
Aymara grammar, as well as to the original data (all words and phrases) captured in the database.


I











1JIJxj


Projects Windows Help
S l | l::1|0 L J-- T IEngish J I1 EDITI
origin

/J I _noun
a note on nouns and verbsw_ bord

more on nouns and verbs verb

definition nouns as verbs or verbs
dnitio 4 |person
definition transforming nouns and pe

into persons in Aymara m eme
I exercise set I lesson definition
sentence suffixes^
defnition entensuxce suffixtypes
riO sentence sufixtypes nterrogative
Information interrogative-.
S4s/ sentence
Sef a note on question marks
d it n es ino interrogative -t euatonal sentence
co personal knowledge suffi.

equational sentences an Term text tyle
S 4,/ Term test 1Itniel
question formation | t| I | u I V I I 0\ MI I
question formation -exam Engish I Spanish I German
sentence construction on. This suffix is used when asking a question to
\ | which you expect only a "yes" or "no" answer. In
other words, all the other information relevant to
end of setl this question is information the speaker already
1 I knows. i



Figure 2. The ObjectEditor is a general purpose tool for creating and browsing database objects. Shown
here is the text description of a portion of the Aymara grammar. There are two different classifications, a
linear student presentation (left side) and a classification by grammatical category (right side).

5.3 Database operations
A significant advantage of ontologies is that they support reasoning facilities which are not available in a
conventional relational database. These facilities can be exploited in linguistic research studies.
Reasoning facilities are based on comparing the structure of two objects to see how they are similar or
different. Reasoners automatically check the consistency of ontology to see that objects are properly
classified. Furthermore, automatic classification can be used to determine where in the taxonomy a new
class or a new individual should be placed. Classification supports many applications, including query
processing. This enables a new way to query databases, based on object structure, that goes well
beyond the industry standard SQL to exploit data semantics in query processing.
Another important reasoning facility is automatic clustering, also known as conceptual clustering,
in which similarities among objects are automatically discovered to form new categories. We can apply
this in the discovery of new phrase patterns by analyzing the similarities over sets of phrases, resulting in
machine learning of grammar. Analysis of a new phrase is accomplished by attempting to apply existing
phrase patterns. Parsing proceeds as a recognition+learning process. First an attempt is made to map
phrase patterns to the new phrase by using a parser (we use a chart parsing algorithm). From another
direction, an attempt is made to match similar phrases to the new phrase to try to discover new patterns.
Since the database contains thousands of phrases (coming directly from the source data corpus) as well
as the abstracted phrase patterns, it is possible to apply case-based reasoning to language learning by
relating the new phrase to this case-base of existing phrases (REF Beck 1994). In our Aymara language
teaching project, we do phrase generation by instantiating phrase patterns (filling in categories appearing
in patterns with individuals that match the phrase pattern constraints).
The ability to reason supports automation of data analysis, leads to machine learning, and
intelligent interaction. While these are useful facilities for studying a single language, they can also be


Z b]trmtr- -risselis










used to study similarities between languages, in particular, since Aymara, Jaqaru and Kawki are closely
related, the database reasoner will be able to automatically identify cross-language similarities among
these languages.

5.4 Data Access, Standards and Issues
We propose to follow standards at the level of data structures for data exchange (XML), rather than at the
level of the particular local software systems being used for implementation. That is, the particular
database management system used to implement our system is not as important as the ability to
exchange the data contained in that database with international partners on other projects using data
exchange standards. Furthermore, the data must also be published in a standardized way, preferably
using a Web Service, so that data can be accessed not only manually by people browsing a Web site
(Such as Figure 1), but by programs needed to access and analyze the data remotely.
The core XML standard used by our system is the Web Ontology Language (OWL), a W3
standard (OWL, 2004). This means that all the objects in our database, including source data,
morphemes, words, phrases, phrase patterns, and dialogs can be accessed in OWL format. Furthermore
we will incorporate the E-MELD GOLD standard as a particular ontology for linguistic categories. We are
not aware of standard Web Service interfaces for the proposed work, but we will collaborate with E-MELD
in the creation of such a standard.
In our analysis of the GOLD standard, we discovered that many of the linguistic categories
needed in our Aymara project, and that will also be needed for the the Jaqaru and Kawki languages, are
not included. Such categories as fourth person, morphological conditioning features, and the morpheme
categories shown in Figure 1 must be added. We will work with the E-MELD community to extend the
standard to support these additional categories. We see GOLD as an evolving standard, and the
introduction of the languages we propose to study provide new insights into categories needed to support
languages internationally.

5.5 Comparison to Work Done Elsewhere (HB)

6. Web Service Architecture, User Interface and Interoperability

There will be two ways to access the database: manually via a Web-based user interface, and
automatically via a Web service that will allow remote programs to attach to and access the database.

6.1 Web Service Architecture

Lyra supports an architecture for publishing the Aymara database on the web in multiple forms. The
database is wrapped inside services that enable the database to be accessed from remote applications.
Primarily the database is wrapped in a web service that supports an application program interface (XML
API) for making calls to the database in order to retrieve data in XML format. Alternatively we also
support a Java RMI (Remote Method Invocation) server that allows remote Java applications to attach to
the database to retrieve data objects directly. The RMI server bypasses the XML API, and that can result
in faster performance for certain types of applications. However, only Java applications can access the
RMI Server (thus it is restricted to that particular language). The LanguageEditor and ObjectEditors are
examples of applications that attach to the RMI Server, and the Aymara training module Flash
environment is an example of an application that uses the XML API.

Figure (6) shows the basic architecture with the linguistic database wrapped inside services, and remote
applications accessing data in various formats. Here are a few of the methods for accessing data
supported by the The XML API:

-getProject(projectlD,language) Gets metadata for a particular project in specified language.
-getUnit(unitlD,language) Gets lesson unit information in specified language.
-getUnitExercises(unitlD,language) Gets exercises for a particular unit.
-getDialog(dialoglD,language) Gets a particular dialog
-getPhraseAnalysis(phraselD,language) Gets syntactic analysis of a particular phrase










getPhrasePatternAnalysis(phrasePatternlD,language) Gets a particular pharse pattern (grammar
rule)
getLesson(lessonlD,language) Gets a particular grammar lesson.
getDictionary(letter, language) Gets dictionary entries beginning with letter.
getDefinition(wordlD,language) Gets dictionary entry for a particular word.

Note while these methods are all for getting information from the database, there are also methods for
submitting data, although those have restricted access for security reasons.




Application 1
VVeb Service (LanguageEditor -
Web I Java)
N
Aymara T Application 2
Language E (Aymara Training
Database R Modules Flash)
Database
E .
Java RMI Server T
XML API Application N





Figur
e 6. The Aymara language database is published on the Internet using various technologies including an XML API
(provides data elements in XML format), and the Java RMI Server (provides data elements as Java objects).
Remote applications can attach to these services to both access and enter data.


6.2 Web-based User Interface
The database can be accessed manually via a standard Web browser equipped with a Java plug-in
(instructions for installing the plug-in are provided). The LanguageEditor in Figure 1 shows an example of
such an interface. This Java applet runs in any browser, and enables users to browse all the database
content (phrase patterns, phrases, word dictionary, morpheme dictionary, etc.). In addition, the applet
provides utilities for data manipulation including automatic classification (which also provides query
processing), conceptual clustering, parsing, generation, and case-based reasoning. The applet also
supports data editing for users who are authorized to do so (usernames and passwords are provided).
Facilities for manually extracting data in XML format will also be available through the Web-based user
interface. Should installation of the Java-plugin be an obstacle for some users (though installation of the
plug-in is not difficult), we will also provide a standard HTML-based interface to as many of the same
functions as possible. Though not as convient as the graphical interface provided by the Java applet, the
HTML-interface is more generic and conventional.

6.3 Web-Service Automatic Remote Access
Because it is also desirable for remote programs to automatically access the data, a Web-service
interface to the database will also be provided. The Web service will enable programs located anywhere
on the Internet and written in any language that can support the Web-service protocol (this includes Java,
C++, C#, PHP and other popular languages) to remotely access and manipulate the data in real time. All
the functions provided in the manual user interface will also be provided via the Web service. Web
services operate by publishing a set of interface calls (methods) via a Web service registry that remote
programs can use to access the service. For example, a method called "Word getWord(String word)"










could be called by a remote application that would retrieve the data object associated with the word
identified in the argument. We will work with the E-MELD community to develop a standard description
language for the Web-service interface.

6.4 Interoperability

While we have developed an OMS comprised of many software systems (authoring tools, visualization
tools, physical storage managers, eLearning) it is not necessary to think of the ontology as physically
residing within a particular database. Rather the ontology is a knowledge network distributed worldwide,
using XML as an exchange format (OWL is XML based and already supports development of distributed
ontologies), with different parts of the ontology managed by different people and organizations. We offer
the OMS as a conceptual framework for achieving that goal. Furthermore, it is important to stress the
significance of data structures over tools. It is no longer necessary to associate knowledge with the
authoring tools used to create that knowledge. Rather the focus should be on the data structures created
by tools. Different tools can operate on the same data structures. The first step towards achieving
interoperability is to change focus from tools to data structures. We also need to distinguish between
archiving standards such as OLAC, which provide course-level metadata descriptions of archived
resources, and database standards that allow sharing of fine-grained data structures. Ultimately of
course it is desirable to share data structures, but the standards are not yet available to achieve that goal.

Interoperability requires standards for data structures. While everyone seems to agree that such
standards are needed, achieving them is difficult. Attempts at standard building include standardizing on
tool-specific data structures [], an interlingua by which proprietary standards are converted to and from an
intermediate accepted standard [], language-to-language converters that attempt to translate between
different data structures without the use of interlingua []. We propose that the ontology, viewed as a
comprehensive, distributed database, can solve the interoperability problem. We argue that other
proposed solutions are basically special cases of this approach. The world-wide linguistics ontology
would be an ever expanding set of data structures and formal definitions (abstractions in the form of
ontology classes) for those structures. The ontology would be physically distributed over many
geographic locations, essential wherever linguistic databases are being built. Communities of Practice
(COP) such as currently envisioned for the GOLD standard [] participate in working on low-level data
structures needed for specific domains while paying attention to related work in the area, and generalizing
their work when appropriate to apply to other areas. Standards building thus becomes a database
building process on a global scale. Most of all it requires a shift in thinking by tool builders who need to
create special data structures to handle the custom features of their tools. Such structures can be
created in the proposed global framework as long as they are registered within an appropriate community
of practice.

6.5 Server Location and Support

The University of Florida's Office of Academic Technologies (OAT) is the central educational computer
support facility at UF and has extensive experience with enterprise server support and secure long-term
data archiving. OAT currently hosts the Lyra ontology management system and associated deployed
applications (this includes EDIS, which receives 6 million visitors per year). The core servers consist of
two dual-processor Linux systems containing 2GB+ main memory and terabyte of disc storage. OAT is
institutionally committed to maintaining the availability of the existing Aymara database and the proposed
new Jaqaru and Kawki linguistic databases over the long term. We will use the servers that currently
support the Aymara database for the proposed project. System administration for the resident servers
(including regular backups, security and support) is provided by the U.F. Office of Academic
Technologies.

7. Work Plan: Roles and Responsibilities

M.J. Hardman, P.I., is the chief researcher for the project and the investigator responsible for all linguistic
matters. She will work closely with project personnel to provide guidance and to write specifications on
the organization of the linguistic database, the selection and parsing of the linguistic and cultural










materials, the translation of the materials, and the involvement of appropriate colleagues in the
assessment of the work. She will carry out the grammatical analysis for the database and act as general
editor for all materials entered into the database. As P.I. Hardman will assure that the project is subject
to appropriate ethical oversight. She will write research papers for presentation at conferences and
publication in journals.
Elizabeth Lowe McCoy, Co-P.I., will serve as the project manager and will oversee the budget, manage
team communications and coordination, and write additional funding documents. She will be the primary
liaison between the University of Florida team and Peruvian counterparts, as well as project liaison with
outside technical bodies such as E-MELD and OLAC. She will work with the content experts, the team
responsible for creating the electronic version of the materials, and the database expert to ensure that the
materials are prepared appropriately for the database, the research function and the online environment.
Dr. Lowe McCoy will oversee English and Spanish language translation of the database materials as well
as supervise the work of the language consultant, Dr. Bautista, to ensure compliance with University of
Florida conflict of interest policies.
Howard Beck, Co-P.I. will be the computer scientist for the project and will ensure the functionality,
robustness and usefulness of the database for research and training purposes. Dr. Beck will also ensure
that the database conforms to current E-MELD and GOLD standards. He will become involved in the
work of E-MELD and write research papers for presentation at conferences and publication in journals.
Sue Legg, Technical Liaison and Evaluation Consultant, will serve as the technical liaison for the
project. In this capacity, she will assist in the planning for the data entry and data analysis functions.
This planning will include the development of procedures and timelines. She will monitor the progress
toward completing these functions and suggest alternatives for overcoming any obstacles. This role also
includes coordinating with the database designer, Howard Beck, and the project directors to facilitate the
resolution of any technical adjustments needed for improving the data entry process. Dr,. Legg will
design evaluation instruments for the project and work with outside evaluators on assessing the use of
the database materials in a variety of contexts.
Dr. Dimas Bautista Iturrizaga, Language Consultant, will be the general arbiter of all the Jaqaru data.
He will be in charge of the writing of definitions for the dictionary and with the country coordinator will
design and conduct training of native Jaqaru and Kawki speakers who will be learning to read and write
their languages, to produce texts and to handle data for the database.
Yolanda Nieves Payano Iturrizaga, Peru Coordinator, will be responsible for recruiting and training
persons in Peru and in Tupe for the entry of material into the database and for using the database for
purposes within the Jaqaru and Kawki speaking communities. She will come to the University of Florida
for short training periods during the life of the grant project.

8. Project Timeline and Procedural Plan

This project will take three years to complete. The project investigators will meet on a weekly basis
during the academic year to assess completed work, monitor progress in relation to project goals, and to
ensure that the project is progressing according to the work plan. The country coordinator will be brought
to the U.F. campus for training and team consultations for eight weeks in the first year and for ongoing
project work for eight weeks in the second year. Ongoing communications with the Peru Coordinator and
her local assistants will occur regularly by phone, email and videoconferencing. The P.I. and the
Language Consultant will travel once a year to Peru to work closely with the Peru Coordinator to provide
support, as well as to monitor the progress of local training efforts and project related work. The work plan
for each year is as follows:

First Year (September 2008 August 2009)
Put team in place; hire graduate and student assistants October 2008
Develop programming specifications for the database
Plan the content of the public facing website November 2008
Digitize notebooks, audio tapes, slides and other materials October 2008-March 2009
Begin data entry into the linguistic database January-August 2009
Begin parsing of the linguistic material January-August 2009
Begin English translations of the linguistic material February-August 2009
Complete missing Spanish translations of the linguistic material March-August 2009










Conduct training for the Peru coordinator at the University of Florida
Catalog and prepare metadata of language resources for AILLA

Second Year (September 2009-August 2010)
Recruit in-country linguists for work on the collaborative dictionary
Conduct training in Peru for Jaqaru and Kawki speakers
Complete data entry of notebooks into the linguistic database
Continue parsing of linguistic material
Continue English translations of the linguistic material
Disseminate the project in Peru and establish positive working ties
With the Peruvian government
Conduct further training and team consultations with Peru coordinator
at the University of Florida
Submit language resources to AILLA


Third Year (September 2010- August 2011)
Complete data entry of audio tapes into the linguistic database
Complete parsing of all the linguistic material
Complete English translation of linguistic material
Edit text files
Edit digitized audio and photo files
Enter photo files into the database
Submit database to the UF Digital Collections
Launch the linguistic database and public website
Present papers at conferences on standards
Conduct a dissemination event in Peru
Solicit funding for follow-on project to create Jaqaru and Kawki
Pedagogical materials


January-March 2009
September 2008-August 2009


September 2009
September 2009-August 2010
September 2009-August 2010
September 2009-August 2010

March 2010-July 2010

January 2010-March 2010

October 2009



September 2010- May 2011
September-December 2010
September -December 2010
September-December 2010
September-December 2010
September-December 2010

January 2011
February-July 2011
April 2011
September 2010 -January 2011


9. Broader Impact and Intellectual Merit

The broader impact of this project will be to preserve and make available in four languages the grammar
and dictionary of two endangered Andean languages for linguistic research and for the use and
collaboration of heritage speakers in a strategically important world region. The intellectual merit of the
project resides in the multifunctional nature of the database, which has features that are distinct from the
SIL LinguaLinks project and the FLEx project, or work done by Charniak and Collins For the Jaqaru and
Kawki speakers, this database of rich text represents a repository of history as well as an opportunity to
contribute to the archiving and preservation of their languages. Some of the recordings are of people
who were born in the 19t" century; they are the words of the grandparents of their grandparents. These
ancestors are no longer with us; their voices constitute the major account of the history of both Tupe and
Cachuy. By making available the language, as well as the notes and images contained in the database,
the proposed project will retrieve and perpetuate the history and culture of the Tupe and Cachuy that will
otherwise be lost. The database materials can be adapted for bilingual education, for literary materials as
well as for historical and cultural purposes, including personal identity affirmation and continuing
education. Training will be provided throughout the life of the grant to enable bilingual and heritage
speakers, as well as those who have lost the language, to use the database for recording their own
material, as well as for pedagogical and research purposes. It is anticipated that these trainees will
become future trainers and teachers of the endangered languages. For the linguistic scholar, having this
type of rich text with full grammatical analysis, in Jaqaru, Kawki, Spanish and English, provides the
resources for comparative linguistics, for the reconstruction of proto-Jaqi, and for adding to the body of
work on linguistic change. Also the existence of this database can make possible additional studies to
further the grammatical analysis of the languages. As examples, the morphophonemics, with some
phonological conditioning but primarily morphological conditioning, is extremely complex, including










noncontiguous conditioning. This database can provide data in a format that will allow further studies of
this system. Also, it is hoped that the study of the distribution of the sentence suffixes in relation to
syntactic structures can be furthered by the use of the database. The nature of the database, together
with the nature of the original transcriptions provides data for future questions that we are not yet asking.
Even after 50 years of research on the Jaqi languages, Hardman has found occasion to ask new
questions, many of which have to do with discourse analysis, sociolinguistics, and dialogue structure.
The existence of this database will allow future studies as new questions are discovered. It will open up
as yet unforeseen possibilities to native speakers of the language as well as linguistic scholars, both in
Peru and around the world.


10. Planned Future Projects

It is hoped that this project will lay the foundation for future work by speakers of Jaqaru and Kawki. In the
future, we would like to bring young scholars to the University of Florida for training in linguistics as well
as computational linguistics so that they can carry on this work in Peru at the university level. The project
investigators would also like to give more direct help to the school teachers in the region who are doing
the difficult work of preserving the language. We hope that the proposed database will help further
research on linguistic databases and will become the foundation for future collaborative work on linguistic
databases for rare and endangered languages.