Finite-State Methods and Natural Language Processing - FSMNLP 2008
Seventh International Workshop
Joint Research Centre of the EC, Ispra, Italy
11-12 September 2008

This year FSMNLP is merged with the FASTAR (Finite Automata Systems - Theoretical and Applied Research) workshop.
AIM and MOTIVATION

The aim of the FSMNLP 2008 is to bring together members of the research and industrial community working on finite-state based models in language technology, computational linguistics, web mining, linguistics, and cognitive science or on related theory and methods in fields such as computer science and mathematics. The workshop will be a forum for researchers and practicioners working

* on NLP applications,
* on the theoretical and implementation aspects, or
* on their combination.

SPECIAL THEME

The special theme of FSMNLP 2008 centers around high performance finite-state devices in large-scale natural language text processing systems and applications. We invite in particular novel high-quality papers related to the topics including:

* practices and experience in deployment of finite-state techniques in real-world applications processing massive amount of natural language data
* industrial-strength finite-state pattern engines for information retrieval, information extraction and related text-mining tasks
* scalability issues in FS-based large-scale text processing systems
* efficient finite-state methods in search engines
* implementation, construction, compression and processing techniques for huge finite-state devices and networks
* novel application and efficiency-oriented finite-state paradigms (compilation and processing), e.g., finite-state devices with rich label annotatations, unification-based finite-state devices
* comparative studies of time and space efficient finite-state methods (vs. other techniques) utilized in NLP applications
* novel appllication areas for finite-state devices in text processing and information management systems
* design patterns for implementing finite-state devices and toolkits

OTHER TOPICS

We also invite submissions that are related to the traditional FSMNLP themes including but not limited to:
1. NLP applications and linguistic aspects of finite-state methods

The topic includes but is not restricted to:

* speech, sign language, phonology, hyphenation, prosody,
* scripts, text normalization, segmentation, tokenization, indexing,
* morphology, stemming, lemmatisation, information retrieval, web mining, spelling correction,
* syntax, POS tagging, partial parsing, disambiguation, information extraction, question answering
* machine translation, translation memories, glossing, dialect adaptation,
* annotated corpora and treebanks, semi-automatic annotation, error mining, searching

2. Finite-state models of language

With this more focused topic (inside 1) we invite papers on aspects that motivate sufficiency of finite-state methods or their subsets for capturing various requirements of natural language processing. The topic includes but is not restricted to:

* performance, linguistic applicability, finite-state hypotheses
* Zipf's law and coverage, model checking against finite corpora
* regular approximations under parameterized complexity, limitations and definitions of relevant complexities such as ambiguity, recursion, crossings, rule applications, constraint violations, reduplication, exponents, discontinuity, path-width, and induction depth
* similarity inferences, dissimilation, segmental length, counter-freeness, asynchronous machines
* garden-path sentences, deterministic parsing, expected parses, Markov chains
* incremental parsing, uncertainty, reliability/variance in stochastic parsing, linear sequential machines

3. Practices for building lexical transducers for the world's languages.

The topic accounts for usability of finite-state methods in NLP. It includes but is not restricted to:

* required user training and consultation, learning curve of non-specialists
* questionnaires, discovery methods, adaptive computer-aided glossing and interlinearization
* example-based grammars, unsupervised learning, semi-automatic learning, user-driven learning (see topic 5 too)
* low literacy level and restricted availability of training data, writing systems/phonology under development, new non-Roman scripts, endangered languages
* linguist's workbenches, stealth-to-wealth parser development
* experiences of using existing tools (e.g. TWOL) for computational morphology and phonology

4. Specification and implementation of sets, relations and multiplicities in NLP using finite state devices

The topic includes but is not restricted to:

* regular rule formalisms, grammar systems, expressions, operations, closure properties, complexities
* algorithms for compilation, approximation, manipulation, optimization, and lazy evaluation of finite machines
* finite string and tree automata, transducers, morphisms and bimorphisms
* weights, registers, multiple tapes, alphabets, state covers and partitions, representations
* locality, constraint propagation, star-free languages, data vs. query complexity
* logical specification, MSO(SLR,matches), FO(Str,(), LTL, generalized restriction, local grammars
* multi-tape automata, same-length relations and partition-based morphology, Semitic morphology
* autosegmental phonology, shuffle, trajectories, synchronization, segmental anchoring, alignment constraints, syllable structure, partial-order reductions
* varieties of regular languages and relations, descriptive complexity of finite-state based grammars
* automaton-based approaches to declarative constraint grammars, constraints in optimality theory
* parallel corpus annotations, register automata, acyclic timed automata

5. Machine learning of finite-state models of natural language

This topic includes but is not restricted to:

* learning regular rule systems, learning topologies of finite automata and transducers
* parameter estimation and smoothing, lexical openness
* computer-driven grammar writing, user-driven grammar learning, discovery procedures
* data scarcity, realistic variations of Gold's model, learnability and cognitive science
* incompletely specified finite-state networks
* model-theoretic grammars, gradient well/ill-formedness

6. Finite-state manipulation software (with relevance to the above themes)

This topic includes but is not restricted to

* regular expression pre-compilers such as regexopt, xfst2fsa, standards and interfaces for finite-state based software components, conversion tools
* tools such as LEXC, Lextools, Intex, XFST, FSM, GRM, WFSC, FIRE Engine, FADD, FSA/UTR, SRILM, FIRE Station and Grail
* free or almost free software such as MIT FST, Carmel, RWTH FSA, FSA Utilities, FSM2.0, Unitex, OpenFIRE, OpenFST, Vaucanson, SFST, PCKIMMO, MONA, Hopskip, ASTL, UCFSM, HaLeX, SML, and WFST (see FSM Registry for more examples)
* results obtainable with such exploration tools as automata, Autographe, Amore, and TESTAS
* visualization tools such as Graphviz and Vaucanson-G
* language-specific resources and descriptions, freely available benchmarking resources

The descriptions of the topics above are not meant to be complete, and should extend to cover all traditional FSMNLP topics. Submitted papers or abstracts may fall in several categories.
SUBMISSION

We expect three kinds of submissions:

* full papers,
* short papers, and
* interactive software demos.

Submissions are electronic and in PDF format via a web-based submission server. Authors are encouraged to use Springer LNCS style (Proceedings and Other Multiauthor Volumes) for LaTeX in producing the PDF document. The page limit for full papers is 12 pages, whereas short papers and software demo descriptions are limited to 6 pages. The information about the author(s) should be omitted in the submitted papers since the review process wil be blind.
PUBLICATION

The papers and abstracts will be published in FSMNLP 2008 proceedings. Publication of revised versions of the papers in a special journal issue is planned.
IMPORTANT DATES

* Paper submissions due: 11 May
* Notification of acceptance: 11 June
* Camera-ready versions due: 30 June