Kirrkirr:Transforming the representation of lexical informationExperiments with endangered language dictionaries Christopher Manning Computer Science and Linguistics, Stanford University (with Jane Simpson, Kevin Jansz, University of Sydney, and Nitin Indurkhya, Nanyang Technological University) http://www.sultry.arts.usyd.edu.au/kirrkirr/
Research Program: Lexicon • A language is more than individual words with a definition • it is a vast network of associations between words and within and across the concepts represented by words • The aim of this work is to provide a wide variety of users – not just linguists – with a better understanding of this conceptual map. • Traditional paper dictionaries offer very limited ways for making such networks visible • On a computer, there are no such limitations to the way information can be displayed.
Research: Computational Lexicography • Dictionaries on computers are now common • But there has been insufficient work trying to utilize the potential of the new medium • Most present a plain, search-oriented representation of the paper version • Goal: fun dictionary tools that are effective for browsing and language learning (cf. Kegl 1995) • Like flicking through a paper dictionary, but better • Innovative ways for representing and linking dictionary information, through creative use of computer software • Should improve user supports and incidental learning • Focus: exploration/dissemination, not creation
Initial focus: Warlpiri • Warlpiri is an Australian Aboriginal language spoken in the Tanami desert (NW of Alice Springs) • There are a number of factors influencing this choice: • Rich lexical materials have been collected by linguists over decades (Ken Hale, MIT, from 1950s, Simpson, Nash, Laughren, Hoogenraad) resulting in the most comprehensive lexical databases for any Australian Language • Warlpiri is the first language of a relatively large community of people. There is reasonable vernacular literacy • Until now, results haven’t been produced in a format usable by the community (only raw printouts) – which is not really acceptable. Fixing this is also good science: for subtle linguistic judgments, one needs speaker involvement.
Educational goals • Dictionary structure and usability are often dictated by professional linguists, while the needs of others (speakers, semi-speakers, young users, second language learners) are not met. Focus: school kids. • The low level of literacy in the region makes an e-dictionary potentially more useful than a paper edition • less dependent on good knowledge of spelling and alphabetical order. • builds on captivating qualities of computers • multimedia content and the pronunciations of words is a considerable help as well.
Kirrkirr: A Warlpiri dictionary browser (Jansz 1998; Jansz, Manning and Indurkhya 1999, 2000) • An environment for the interactive exploration of dictionaries. • Although our current work has just been with Warlpiri, the design is general – any XML dictionary • Attempts to more fully utilise graphical interfaces, hypertext, multimedia, and different ways of indexing and accessing information • Written in Java, it can either be run over the web (needs bandwidth) or locally (here Java’s main advantage is cross-platform support: Win/Mac/Unix) • originally JDK1.1.6+Swing, now Java 2
Overview Kirrkirr provides various modules • Animated network layout of word relationships • Formatted dictionary entries • A notes facility for ‘jotting in the margin’ annotations • Multimedia: audio, pictures • Semantic domain browsing • Advanced searching interfaces • others in planning: formatting (XSL) editing, figuration patterns, terminology sets • These attempt to cater to users with different interests and competence levels
The lexical database • Original text materials are stored in an ad hoc format of markup using backslash codes with some (rather odd) nesting of structural tags [origin: troff] • These are converted to XML using an error-correcting stack-based parser (written in PERL) • The inconsistency and flexibility of dictionary entries actually made this a surprisingly difficult task. • Innumerable structural errors/inconsistencies/typos from years of hand maintenance in text editors and via regexps • Innumerable problems with link consistency • Heuristic content-sensitive parser imposes data integrity • After much grief – Software Engineering 101 – we now have a ‘one click’ process for regenerating the whole system
Data representation: XML • Use of XML gives a clear structure to the lexical data • Separates structure of the data from its presentation • Much of the recent enthusiasm for XML has involved textual representation of simple and rigid structures well-represented in RDB records (e-commerce, etc.) • But dictionary entries thoroughly exploit XML • rich hierarchical structure • entities vary greatly depending on the word being defined • Result remains a portable, tangible text file Use of standards for field linguistic data means: many (free) tools are available, extra functionality comes for free, one can interoperate with mainstream software.
Kirrkirr’s XML Index Process Index in Memory XML Formatted Warlpiri dictionary file word file position word file position word file position <DICTIONARY> <ENTRY> ... </ENTRY> <ENTRY> ... </ENTRY> <ENTRY> ... </ENTRY> </DICTIONARY> Across file system or web XML Parser Kirrkirr Dictionary Browser XML Document Object Model We currently uses ad hoc indexing of the XML for efficient access (but expect to move to XQL, as it matures).
Visualization of dictionary information • For dictionaries with simple textual content behind them, there is little that can be done but an on-line reflection of a printed page • But we would like to be able to do more • we want to know a word’s relationships to other words, and the patterning in these relationships • In a computational approach, the program can mediate between lexical data and the user • The interface can select from and choose how to present information (according to the user’s preferences and abilities) – in many different ways
Graph-based visualisation (Jansz 1998; Jansz, Manning and Indurkhya 1999) • Classic graph layout problem • Adapts work by Eades et al. (1998) and Huang et al. (1998) on visualisation and navigation of WWW document linkages • Uses the spring algorithm. Big advantage is that it is an iterative updating algorithm, and so gives an easy interactivity: • it wiggles and people can play with it, clicking to sprout nodes • A major goal was clarity and simplicity of the graph: the software maintains a set of focus nodes to prevent overcrowding
Formatted dictionary entries • Are produced automatically and online from the XML by using XSL(T) – a tree transformation language • XSL allows easy modelling of some user preferences • One can leave out information such as part of speech, or detailed definitions, or rearrange it • We provide several stylesheets to choose from • This issue is surprisingly important: many users find information overload confusing and demotivating • Can produce a bilingual or monolingual dictionary • Can also use this for print dictionaries (via RTF or TeX). We have produced a couple of samples.
Rich typology of link types • The semantic links present in a dictionary (synonym, antonym, hyponym, subentry, variant, coverbs, …) solve a major problem of the web: we have many link types each with a clear semantic interpretation • We use consistent colour-coding of text and network edges to show these link types • Gives a richer browsing experience • You can tell where you are going before clicking • Dictionary-given links are supplemented by links derived from collocational analysis of Warlpiri texts • uses loglikelihood ratios (Dunning 1993) • works reasonably successfully from 1/4 million words
Educational advantages/usability • Work (at PARC and elsewhere: Pirolli et al. 1996) has stressed the role for browsing as well as searching in information access • It provides a context for learning • A student can opportunistically explore words that are related in various ways • Important semantic relationships can be understood • People continually see alphabetical order and word spellings, but don’t need to know them to use Kirrkirr • Use of “fuzzy spelling” in searches supports users with poor spelling. It usually finds what you wanted.
Multimedia (currently pictures and audio) Can hear pronunciations – gives a much better understanding of pronunciation than phonetic symbols pictures of plants and animals are more intelligible than descriptions (future: videos of Warlpiri sign language …) Advanced search page search various fields, regular expressions, fuzzy spelling, etc. Notes: one can annotate dictionary entries (to correct or personalise) Other components
User study Mim Corris (Yuendumu, Willowra), Jane Simpson (Lajamanu) • User testing with primary and (lower) secondary students (observation, and dictionary tasks) • Observation of trainee Warlpiri literacy workers • Comments from teachers, other adults, etc. • Qualitative ethnographic study of dictionary use. (Doing anything much else would be difficult.) • Initial reactions are very enthusiastic • Students used it voluntarily during lunch breaks • Could use as a basis for classroom activities (better with some further development: games and puzzles)
A positive anecdote “One of the introductory Warlpiri literacy students, who had not been very interested in the literacy class, spent nearly 3/4 hour looking at Kirrkirr apparently in absorbed concentration. She wasn’t especially interested in the sound and picture possibilities. She moved between words, scrolling along the list, typing in the search, clicking on the words in the network pane. She wasn’t even put off when the dictionary definitions stopped appearing – looking at the networks of words instead. … After the Kirrkirr demo she asked if she could have a printed dictionary to take away with her to use in camp to learn the words. I interpret this as a desire to learn words in her own time and place.”
Endangered language dictionaries (Corris, Manning, Poetsch, and Simpson 1999). Based on 72 people. • We’ve been testing both paper and electronic dictionary desires, use and usability • competing goals: documentation dictionaries vs. maintenance/learning dictionaries (linguist vs. other user) • symbolic organization vs. practically useful organization • lack of understanding of dictionaries, and limited literacy can make paper dictionaries ineffective • 45–60 minutes for 12 dictionary lookups… • lack of electricity makes e-dictionaries ineffective in some places (e.g., Indonesia – but OK in Australian schools) • E-dictionaries can solve many usability issues • font size, amount of info, ‘infinite’ space, easy lookup, sound, customizability
Interim Conclusions • Kirrkirr is a prototype of what one can do to develop new ways to organize and visualize lexicons • We have addressed the challenge of making dictionary information accessible and usable in the creation of an application which mediates between well-structured data and users’ needs and insights in searching/browsing and presentation • It allows web distribution of information from a server • We’ve begun making it available in Warlpiri schools – more results to follow.
Kirrkirr: Experiences with a flexible software interface to indigenous dictionaries Christopher Manning Computer Science and Linguistics, Stanford University (with Kevin Jansz, University of Sydney, and Nitin Indurkhya, Nanyang Technological University) http://www.sultry.arts.usyd.edu.au/kirrkirr/