Paper Title (use style: paper title)
Contents
1. entry in the lexicon is represented as a DT structure Among the information that represents an entry lemma in a GRAALAN lexicon we find the following information involved in inflection process the lemma itself in normal and phonetic alphabet the lemma syllabification in normal and phonetic alphabet eventually the morphological syllabification the lemma AVT the structure DT for the multiword lemmas and the inflection rule label that must be applied to the lemma in order to obtain all lemma extended paradigm The current GRAALAN Romanian language lexicon not yet finished contains about 76 000 lemmas 66 000 mono word lemma 10 000 multiword lemmas 115 000 senses 103 000 supplements and 12 700 multiword expressions VI THE SYLLABIFICATION RULES The inflected process in GRAALAN system is accomplished in parallel with the inflected form syllabification The rules for the syllabification are described in a special GRAALAN section There are three types of syllabification considered by GRAALAN a The normal or euphonic syllabification that refers to normal alphabet written words b The phonetic syllabification that refers to the phonetic alphabet written words c The morphological syllabification that refers to the normal alphabet written words but also considers some morphological criteria 9 In fact only the first two syllabifications have syllabification rules and these rules must be described in the cor2. B b alphabet 2 phonetic 36 open_central_unrounded alphabet mid_central_unrounded mid_front_unrounded 3 special 64 tilde grave_accent commercial_at characters 4 stress markers 2 primary_stress secondary_stress 5 groups 360 a_group tch_group sh_group 6 classes 17 capital_letter small_letter vowel semivowel consonant diphthong Additional information for different types of characters can be specified codes names sorting sequence direction of writing to the left to the right some special functions etc V THE LEXICON GRAALAN lexicon is a very complex data structure much more complex than a usual lexicon We will mention here only few of the lexicon features those that are involved in the inflection process A GRAALAN lexicon is a section with a large number of entries There are three types of entries lexical entries morphological entries end procedural entries a Lexical entries contain morphemes words and multiword expressions MWE Morphemes are particles that contribute to word formation roots prefixes suffixes prefixoids suffixoids prime elements median elements final elements Words are the basic elements of the syntactic constructs lemmas basic or canonical word forms for example the nominative singular forms for nouns supplements for example the nominative plural forms for nouns different other word forms from the word paradigm for example the second person of
3. Central word Text actor Phonetic akt or articulare nearticulat numar singular caz nominativ gen masculin animatie animat tip substantiv comun clasa substantiv Auxiliary words Text un Phonetic un Reference Art01 clasa articol tip articol nehotarat caz nominativ gen masculin numar singular Belongs yes acord art end of entry The current implementation for Romanian language generates 17 928 056 inflection situations 2 075 978 mono words and 15 852 078 multiwords and 9 946 686 inflection forms 1 019 783 mono words and 8 926 903 multiwords IX INFLECTION EXECUTION The inflection process is realised by a special tool named MKT MKT had to be designed not only to automatically generate the inflection situations inflected forms but to help the linguist to correct inflected forms of a particular lemma to save and to generate new inflection rules according to changes made by the linguist The system should allow also the linguist to check the inflected forms previously saved The actions that a linguist can apply to generated inflection forms are editing deleting and inserting Other possible actions are creating syllabification forms and choosing values for not inflectional attributes for the purpose of generating the entire inflection situation tree for the current lemma In order to decrease the entire correction of inflection forms time the system generates and dis
4. forms the linguist can generate new syllabification forms MKT system can run in debug mode for inflection rules which may help detecting some errors During runtime MKT generates fully detailed HTML files for each basic rule These reports show basic rules written in GRAALAN language with additional information about all elementary rules variants conditions actions and a complete log for all lemmas that used them After generating inflected forms MKT adds them to the inflected form LKB Modifying inflected forms automatic generated by the inflection rules or adding other forms involves generating new basic rules grouped in a new compound which will be referenced by the lemma The consequences of adjustments made by the linguists are almost transparent for them because MKT automatic determines the minimum number of steps for saving the information and it acts with no user help If the linguist corrects only the syllabification forms then the system will save only those forms If the linguist corrects the automated generated inflected forms then MKT will update inflected forms and inflection rules too If the linguist modifies deletes or inserts a new inflected form for an inflection situation which has a correspondent inflection situation in one of the active basic rule for that rule will be generated a new version specific to the current lemma and according to the changes made by the linguist If the linguist inserts inflected for
5. inflection forms are obtained by applying the inflection rules on lemmas from the lexicon A collection of inflection forms is a set of entries An entry corresponds to an inflected form An inflected form can have one or many words When an inflected form has many words one of these words is considered center and it will be used to do eventually an inflection forms tri An entry contains The characterisation of the inflection situation situations of the inflected form i e an AVT Therefore we group more inflection situations that have associated the same inflected form The inflection form in normal and in phonetic alphabet The label of the word from the lexicon where we find the lemma of the current inflection form The normal phonetic and eventually the morphologic inflection form and the syllabification at the end of a line the hyphenation The indication of the word used for tri in the case of multiword inflection forms The structure DT in the multiword inflected form A sample of an entry in inflection form s list for Romanian language is the following ETF_Entry_1 1 1_1_1 Entry Text un actor Phonetic un akt or Reference Entry_1 1 1_1 clasa substantiv tip substantiv comun animatie animat gen masculin numar singular caz nominativ acuzativ articulare nehotarat wow Syllabification Euphonic un ac tor ak t or Tri 1 left mi Phonetic un
6. is not a language of pure type the characteristics of different types being combined at different degrees For example Romanian and French languages combine synthetic and analytic features yet Romanian language is more synthetic than French language We will consider here that a word is the basic unit of vocabulary and represents an association between a sense and a sign or a string of signs The sign can be a sound sign graphic sign or any other type of sign GRAALAN system 8 implements GRAALAN language Grammar Abstract Language 7 and its purpose is to allow the description of certain linguistic aspects regarding one natural language or the correspondences between two natural languages In GRAALAN we will consider the graphic signs coded according to UNICODE 12 and the sound signs which are also represented by some graphic signs according to IPA International Phonetic Alphabet 10 for example that is also UNICODE codified GRAALAN system is more appropriate for the non insulated inflected synthetic or analytic languages but it can also be used for other types of languages When a word is used in a phrase it can be in different inflection situations according to the syntactic role that the word plays in the phrase In GRAALAN an inflected situation is an ensemble of morphological attributes and values of these attributes organised as an AVT Attribute Value Tree 6 An inflected form corresponds to each infle
7. the indication if the value is applied to the lemma the name of a function that procedurally makes the treatment if it is the case etc For the morphological configuration implemented for Romanian language we established a variant with see Table D 11 basic classes noun article pronoun adjective numeral verb adverb preposition conjunction interjection and a special class named sign 819 attribute apparitions among these there are 111 inflected attribute apparitions using 67 distinct attribute names and 2 122 attribute values apparitions using 211 distinct attribute value names This morphological configurator defines 19 462 ECs the verb of course has the most ECs 16 320 The maximum EC length is 36 We must say that the description in a morphological configurator is not unique It can be accomplished using more or less attributes and attribute values according to the target of the research This implies that the number of inflected situations will be different depending on the number of attributes and attribute values that are used TABLE I ROMANIAN MORPHOLOGICAL CONFIGURATOR Total Inflected Total EC Maximum Attributes Values Name attributes attributes values Number EC length names names Class 819 111 2 122 19 462 36 67 211 Class Noun 9 3 27 312 12 6 18 Class Article 13 2 38 82 8 4 15 Class Adjective 12 0 30 420 14 9 21 Class Pronoun 112 12 324 1 118 16 13 45 Class
8. the personal noun Multiword expressions are groups of words that have a global sense very different from the senses of each single word In GRAALAN lexicon these MWEs are represented not only as a word sequence but in a more elaborated form using a dependency tree DT b Morphological entries contain information about the language morphology and are correlated with morphological configurator morphological categories and values synthetic characteristics for the mono word inflected forms and analytical characteristics for the multiword inflected forms c Procedural entries describe the procedures that are eventually called in different parts of the linguistic description From the inflection point of view the most important type of entry is the lexical lemma A lemma is in fact in GRAALAN one of the inflected forms associated to an inflection situation from the extended paradigm of a word used in a conventional mode as a reference form for the others inflected forms A lemma can eventually contain more words for example a zbura to fly in Romanian language One of the words of such a multiword lemma is considered as inflected form s center and it is used in the sorting process The other words of the inflected form are considered auxiliary words Each auxiliary word will be accompanied by different information references to lemma the auxiliary word originates from an associated AVT Consequently a multiword lemma
9. GENERAL SYSTEM FOR NORMAL AND PHONETIC INFLECTION Stefan DIACONESCU Cristi INGINERU Felicia CODIRLASU Monica RIZEA Oana BULIBASA gt Department of Research and Development SOFTWIN Bucharest Romania sdiaconescu softwin ro Abstract Generating all inflected forms for a natural language is a very difficult task not only because of the large number of inflection rules but also because of the large number of exceptions The operation is more difficult if we consider not only the normal inflected forms where the inflected forms are represented in the normal alphabet of the involved language but also the phonetic forms where the inflected forms are represented in a phonetic alphabet This paper presents a general system i e a system that can be used for any inflected natural language that generates all the normal and phonetic forms of words from a given lexicon A set of results obtained by using this system for Romanian language is also presented The system is based on GRAALAN metalanguage Grammar Abstract Language used for representing the linguistic knowledge concerning a natural language and requested by the inflection process It also uses a set of tools that allows the handling of this meta language The system starts with a set of linguistic knowledge bases described in GRAALAN containing normal and phonetic inflection rules the description of the normal and phonetic alphabets that are used the base forms lemmas th
10. Numeral 154 43 447 1 124 16 20 63 Class Verb 478 35 1 142 16 320 34 24 73 Class Adverb 33 8 89 68 14 12 32 Class Preposition 1 1 4 3 2 1 4 Class Conjunction 3 3 10 7 4 3 10 Class Interjection 2 2 5 3 4 2 5 IV THE ALPHABET In GRAALAN by the alphabet of a language we understand the set of signs usual alphabetic signs syllabic signs and ideographic signs with their codes that are used to write texts and to describe phonetic aspects of that language In a GRAALAN alphabet section the following subsections are presented a The normal alphabet is the set of signs used to write texts in the language b The phonetic alphabet is a set of signs used to phonetic transcription of the language sounds c The special characters are other signs used for example punctuation marks d Stress markers specify the primary stress and the secondary stress e Character groups like diphthongs and trifthongs for Romanian language indicate the relation between the normal written signs and the phonetic written signs f Classes define the names for different categories of signs for example the consonant class the vowel class the capital letters etc Table II shows the number of alphabet signs used for Romanian language in the present implementation TABLE II GRAALAN SIGNS FOR ROMANIAN LANGUAGE No Sign types Signs Examples of sign names number 1 normal 66 A a
11. X 1 2 User Manual Universit de Marne la Vall e 2006 The Penguin English Dictionary Editura Litera International copyright Merriam Webster Inc 2003
12. at will be inflected taken from a lexicon also in normal and phonetic forms and the description of the morphological structure of the language morphological categories and their values All these linguistic knowledge bases are created using special tools The inflected normal phonetic forms will be automatically generated by applying the normal phonetic inflection rules on the lemmas of the lexicon taking into account the morphological structure and the normal phonetic alphabets At the end of the process the linguist is able to verify and eventually correct the generated forms When corrections are made new inflection rules are automatically generated The presented system is part of larger system for natural language processing Keywords component inflection rules phonetic inflection rules inflected forms natural language processing I INTRODUCTION HEADING 1 Though it seems to be quite peculiar there is not a generally accepted definition of the word notion According to 10 the word is the basic unit of the vocabulary that represents the association of a sense or of a complex of senses and a sound complex In 15 it is given a more general definition meaningful unit of spoken language that can stand alone as utterance and is not divisible into units and also a more restrictive one written or printed representation of a spoken word that is usually set off by spaces on either side In 3 there is a general defi
13. ction situation Many inflection situations can have the same inflected form in some cases for non inflected words or for insulated languages the inflected form is unique therefore it loses attribute inflected The set of all inflection situations of a word is named in GRAALAN word extended paradigm Usually a word paradigm is defined in a more restrictive way as the set of inflected forms The larger definition in GRAALAN allows considering the non insulated languages using the same formalism Consequently even languages with very poor inflection like English or almost no inflection like Chinese can be characterized by a quite large extended word paradigm Word inflection is represented in GRAALAN as the process of obtaining all the inflection situations of a word together with all the corresponding written inflected forms This process refers not only to the normal alphabet written word we understand here by alphabet the usual alphabet and also syllabic or ideographic transcription but also to the phonetic alphabet written word Unlike the other automatic inflection systems 2 14 automatic phonetisation systems l and automatic syllabification systems 11 that usually produce different kinds of dictionaries specific to a certain language and function GRAALAN system is a unitary system that groups all the above functions and many others and gives them a structure as part of a larger NLP Natural Languages Proc
14. dths line spacing and type styles are built in examples of the type styles are provided throughout this document and are identified in italic type within parentheses following the example Some components such as multi leveled equations graphics and tables are not prescribed although the various table text styles are provided The formatter will need to create these components incorporating the applicable criteria that follow II THE PLACE OF INFLECTION SYSTEM IN A NLP SYSTEM GRAALAN language allows the description of the following linguistic sections alphabet in a very large sense morphological configurator syllabification rules lexicon inflection rules inflection forms syntax rules correspondence between two natural languages The inflection process uses only the following sections alphabet morphological configurator syllabification rules lexicon inflection rules and inflection forms see Fig 1 Actually the syllabification does not belong to inflection process but it is completed in the same time and with the same tool as the inflection Each of the five input GRAALAN sections must be described by the linguist for a certain natural language see Fig 2 using GRAALAN language and tools The GRAALAN text is compiled using a GRAALAN compiler and transformed in a knowledge base LKB corresponding to each section which is stored in XML Extensible Markup Language format 13 The lexicon section can be also
15. e Informatique Statistique Socio linguistique I COTEANU L SECHE M SECHE et al DEX Dic ionar Explicativ al Limbii Rom ne Editura Univers Enciclopedic 1998 S DIACONESCU Some Properties of the Attribute Value Trees Used for Linguistic Knowledge Reoresentation in 2nd Indian International Conference on Artificial Intelligence IICAI 05 India 2005 S DIACONESCU Natural Language Understanding Using Generative Dependency Grammars in Max Bramer Alun Preece and Frans Coenen Eds ES 2002 Research and Development in Intelligent Systems XIX Proceedings of ES2002 the Twenty second SGAI International Conference on Knowledge Based Systems and Applied Artificial Intelligence Cambridge UK December 2002 Springer pp 439 452 S DIACONESCU Complex Natural Language Processing System Architecture in Corneliu Burileanu Horia Nicolai Teodorescu Eds Advances in Spoken Language Technology The Publishing House of the Romanian Academy Bucharest 2007 pp 228 240 M St ILINCA Gramatica Instrumentala Editura Festina 1995 International Phonetic Association Handbook of the International Phonetic Alphabet Cambridge University Press 2005 Y MARCHAND C R ADESTT R I DAMPER Automatic Syllabification in English A Comparison of Different Algorithms 2008 The UNICODE Consortium The UNICODE Standard Version 5 0 Fifth Edition Addison Wesley Professional 2006 W3C Extensible Markup Language XML 1 0 Recommandacion 1998 UNITE
16. es b Loading inflection forms only if those were generated and saved before c Selecting values for not inflectional attributes if the current lemma s inflection situation is not enough to build the exact word tree the inflection situations of saved inflection forms are being used to choose values for not inflectional attributes If there were not previously saved inflection forms some information has to be supplied by the linguist using the graphic interface d Generating inflection form nodes for all inflection situations from the word tree the system searches for the corresponding elementary inflection rules For all elementary rules attached to the inflection situation found in basic rules new inflection form nodes are being created and added to the word tree e The generation of the inflection forms associated to inflection form nodes The generation consists on traversing the oriented graph of elementary rules Because an inflection form can be obtained sometimes from another inflected form a loop can appear A checking on this graph assures that there are no cycles that could put the program into an infinite loop f Syllabification If there are previously saved inflected forms the system tries to attach them to newly generated inflected forms comparing inflection situations and alphabetic phonetic text If the previously saved inflected form is not correct or if there are no previously saved inflected
17. essing system Section 2 will present how an inflected process takes place in a GRAALAN based processing system The inflected process uses some previously created linguistic knowledge bases morphological configurator section 3 alphabet section 4 lexicon section 5 syllabification rules section 6 and inflection rules section 7 The inflection situations and the inflected forms that are generated during the inflection process are presented in section 8 The application MKT Morphological Knowledge Tool that accomplishes the inflected process itself is presented in section 9 For each of sections 3 to 8 some effective implementation for Romanian language is presented too In section 10 some conclusions about the generality of GRAALAN system language and the usability in different linguistic applications are presented This template modified in MS Word 2003 and saved as Word 97 2003 amp 6 0 95 RTF for the PC provides authors with most of the formatting specifications needed for preparing Alphabet Morph Config Inflection rules Inflection system electronic versions of their papers All standard paper components have been specified for three reasons 1 ease of use when formatting individual papers 2 automatic compliance to electronic requirements that facilitate the concurrent or later production of electronic products and 3 conformity of style throughout a conference proceedings Margins column wi
18. flection situation and inflection rule There are two kinds of steps for generating inflection forms the word tree preparation and the inflecting itself I The word tree preparation It is made only once after the linguist has been selected the lemma It works for more lemmas in the same way that it works for only one The input of this section is the inflection situation and the morphologic configurator The result is the word tree which later will be filled with inflected forms a Copying morphologic configurator a new copy of morphologic configurator is saved and the original one is kept for further usages b Loading lemma s inflection situation the system loads the inflection situation of current lemma or lemmas from the lexicon c Intersection the morphologic configurator copy is filtered using the inflection situation of the lemmas from the lexicon The result is the word tree d Expanding the word tree is being expanded which means that the attributes will have each at most one value Il The inflecting During a work session the user can modify some automatic generated inflected forms He can eventually cancel the modifications made by himself and command the new automatic inflected form generation etc The following steps are selected according to the user actions a Loading inflection rule this step involves the reading of the inflection compound rule and the associated basic inflection tul
19. introduced using a special tool LKT Lexicon Knowledge Tool that automatically generates GRAALAN text The lexicon GRAALAN text is compiled and then the lexicon XML LKB is generated Afterwards the five XML LKBs are used by MKT which implements the inflected normal and phonetic process and generates inflected forms in GRAALAN Then the inflected forms are compiled and transformed in a XML LKB Fig 1 MKT is a tool that allows the linguist to also verify and to correct the inflected forms using an interactive interface Should some modifications be necessary caused for example by some exceptions not yet considered MKT will automatically generate in GRAALAN new inflection rules in order to specify this modifications The new rules will be compiled and the XML LKB with inflected forms is updated The knowledge bases containing inflected forms will be then used by different tools and applications as showed in Fig 2 Figure 1 GRAALAN Sections involved in Inflection System Linguist Alphabet GRAALAN Morph Config GRAALAN Syllab rules GRAALAN Inflection rules GRAALAN Lexicon GRAALAN Inflection forms GRAALAN HI THE MORPHOLOGICAL CONFIGURATOR The morphological configurator of a natural language represents the ensemble of morphological categories e g class gender number case etc and values of these morphological categories for example noun adjective verb etc for cla
20. l the words from the lexicon can have For a not inflected word we will have only one inflection situation In Fig 3 an extremely simplified AVT is represented It could also be represented as follows class noun gender masculine feminine neuter number singular plural It will result eight ECs and many inflection situations class noun gender masculine number singular class noun gender masculine number plural Il I class noun gender feminine number singular I Il class noun gender feminine number plural class noun gender neuter number singular class noun gender neuter number plural class verb number singular N N class N A verb me gt 1 1 class verb number plural masculine gender feminine neuter singular number plural singular number pural Figure 3 Example of a simplified AVT In GRAALAN each node has associated not only the attribute name or the attribute value name but also some other information a For attributes the abbreviation of the name the indication if the morphological category is inflected or not the name of a function that procedurally makes the treatment if it is the case GRAALAN is a declarative language but it can call some procedural function if it is needed b For attribute values the abbreviation of the name
21. lp remove the unused basic and compound rules or the equivalents among them The results of this function are transparent for the linguist and the function takes care about keeping valid references among all lemmas compound basic and elementary rules Deure Select lemmas rex substcommasc H Rules Filter Inflect Lou E clasa E substantiv amp tip substantiv Sle comun El animatie E animat Fe gen E e masculin numar B singular B caz E nominativ B articulare El nearticulat gt Bi problematic problem atik pro ble majtic pro ble m a tik F problemist problem ist pro ble mist pro ble m ist gt BBolacheu plac eu pla che u pla c e u deputat deput at de pu tat de pu t at nehotarat s Bn problematic un problem atik un pro ble ma tic un pro ble m a tik un deputat un deput at un de pu tat un de pu t at hotarat gt Boroblematicul problem atikul pro ble ma ti cul pro ble m a ti kul F problernistul problem istul pro ble mis tul pro ble m is tul Ee acuzativ H articulare El nearticulat me problemist un problem ist un pro ble mist un pro ble r ist BBolacheut plac eul pla che ul pla c e ul deputatul deput atul de pu ta tul de pu t a tul Form status Actions successfully applied p Inflection Form Alphabetic un placheu Phonetic un plac eu r Syllabification A
22. lphabetical un pla che u Phonetical un pla c e u Morphological Reset Preview prandsyi AuxDeser Variant ERule l Text placheu Phonetic plac eu clasa substantiv animatie animat tip substantiv comun gen masculin numar singular caz nominativ articulare nearticulat T Auto Save Auxiliary Word Text un Phonetic un Reference Art0 clasa articol tip articol nehotarat caz nominativ gen masculin numar singular Belongs yes rel art nehotarat H Bbproblematic problem atik pro ble ma tic pro ble m a tik f clasa substantiv tip substantiv comun animatie animat gen masculin numar singular caz nominativ Position 7 Total 112 Filtered 112 Ready ks 8286 placheu gt gt m z Figure 5 MKT Graphical user interface X CONCLUSIONS The normal and phonetic inflection forms generation process according to GRAALAN language has a great generalisation degree It can be applied to almost all the natural languages and is particularly efficient for not insulated inflected languages If the used linguistic knowledge bases alphabet morphological configurator lexicon syllabification rules inflection rules are complete from the inflection process point of view this process is automatic Anyway it is advisable to verify the results in order to see if there are some errors in the input LKB and to see if some exce
23. mma from the lexicon indicates a compound inflection rule that is applied in order to obtain the extended word paradigm A compound inflection rule is a list of basic inflection rules A basic inflection rule is formed by a set of inflection situations that can be generated by the current basic inflection rule represented as an AVT and a set of elementary inflection rules An elementary inflection rule contains a reference to a lemma or to another inflection situation and a list of variants The transformations needed by the inflection process will be executed on the inflected form associated to this referred inflection situation One variant from the elementary inflection rule contains a condition a set of normal transformations a set of phonetic transformations and a structure DT and AVT for multiword inflection forms The condition is an expression with a logical value yes no that has as operands lemmas or other elements from the current word paradigm normal or phonetic character strings groups or classes see alphabet section and as operators the identity operators The normal transformation is an operation insert replace or delete executed on normal representation of the inflection form associated to the referred inflection situation The phonetic transformation is an operation insert replace or delete executed on phonetic representation of the inflection form associated to the referred inflection situation Elementar
24. ms for at least one inflection situation with no correspondent in the active basic rules then a new basic rule will be generates having the shape of the AVT resulted from the union of all inflection situation with no correspondent in the active basic rules but with at least one inflected form added by the linguist Generating new basic inflection rule involves generating new elementary rules with generated transformations lists and replacing the old elementary rules The newly created elementary rules have one variant each they reference the lemma text and they have no associated condition The generation of lists with actions for transforming the alphabetic and phonetic text of lemma into alphabetic and phonetic text of inflection form uses the Levenshtein algorithm This algorithm shows how a minimum list of actions like deleting inserting and replacing can be obtained to convert one given string into another given string 14 If there was a new basic rule generated from scratch or generated as a new version of another basic rule then the list of all basic rules used by the current lemma will be found into a generated compound rule and the current lemma will reference not the old one but the new compound rule At the end of work session a special function can be called to clean up the knowledge base The main purpose of this function is to Palirom Database INDEX xml cfg E3 Eile Edit View Selection Function Options Window He
25. nition the complex linguistic unit simultaneous realised as phonetic semantic and grammatical unit and one that refers to written words characters group in an alphabetic syllabic ideograph transcription set off by typographical white spaces In the general case different structures of a word must be considered phonological morphologic semantic and syntactic structures All these definitions seem to be more appropriate for indo european languages The notion of word must be more refined for agglutinant or insulated languages Consequently the notion of word is finally dependent on the considered natural language By a typological classification there are two types of natural languages insulated like Chinese or non insulated In the insulated languages words are generally invariable the different situations of a word being represented by syntactic elements that consider the word order for example and intonation The non insulated languages can be agglutinant or inflected In agglutinant languages like Turkish words are formed by a root and a number of affixes Sometimes it is not very easy to make the difference between agglutinant and inflected characteristics The inflected languages can be grouped into synthetic and analytic languages In synthetic languages words are formed by many morphemes that fuse together The analytic languages use syntactic means in order to obtain some situations of words Actually there
26. plays those inflection forms of more than one lemma in the same time This facility saves some dead times of loading rules also saving some other various actions that the linguist can take for all inflected forms of more than one lemma The only condition for simultaneously working with more lemmas is that all lemmas must have the same inflection situation and the same compound rule All the inflection situations with the corresponding inflected forms of a particular lemma constitute the word tree see Fig 5 The word tree is obtained from morphologic configurator filtered on the inflection situation of a lemma The filtering process consists in deleting the undesired values of not inflectional attributes according to the value of attributes from the lemma s inflection situation For example if the lemma s inflection situation says that it is about a verb then all other lexical classes are ignored A basic inflection rule is valid only in a special context when it is included in a compound rule beside other basic rules because all elementary rule labels must be unique and all references must be solved The program had been limited to use only an active compound rule at a given time Active compound rule means that all its refered basic rules are found in memory and the elementary rule label references have been searched and indexed for a fast access during runtime In conclusion from the MKT point of view lemmas can be grouped by in
27. ptions were not omitted especially in the inflection rules MKT offers to this purpose a graphical user interface The set of generated linguistic knowledge bases can be used in other types of application for example starting with the simplest ones like spellers lemmatisers that use only mono word inflection forms mono word annotators that use the mono word inflection forms but also the inflection situations up to more complex applications like grammar checkers multiword annotators that need information from the other GRAALAN knowledge bases too Finally if we dispose of the description of many natural languages the GRAALAN correspondences between pair of languages can be described and after that used in automatic or assisted machine translation REFERENCES 1 V AUBERGE R BELRHALI La phon tisation automatique du fran ais mergence de r gles ou de lexiques Institut de la communication parl e Grenoble FRANCE 1991 2 A M BARBU Romanian Lexical Data Bases Inflected and Syllabic Forms Dictionaries The 6th edition of the Language Resources and Evaluation Conference 2008 3 A BIDU VR NCEANU C C L RA U L IONESCU RUXANDOIU M MANCAS G PANA DINDELEGAN Dic ionar de tiin e ale limbii Editura Nemira 2005 4 5 6 7 8 9 10 11 12 13 14 15 C CHARRAS T LECROQ Sequence comparison LIR Laboratoire d Informatique de Rouen and ABISS Atelier Biologi
28. responding GRAALAN section An example of normal syllabification is the following Rule amp vowel amp semivowel amp semivowel amp semivowel amp vowel By this notation the sign means must not be separated and the sign means it can be separated The amp vowel amp semivowel amp semivowel amp semivowel amp vowel noted characters according to XML conventions refer to normal characters The corresponding sample for phonetic syllabification is the following Rule amp phon_vowel amp phon_semivowel amp phon_semivowel amp phon_semivowel amp phon_vowel The amp phon_vowel amp phon_semivowel amp phon_semivowel amp phon_semivowel amp phon_vowel noted characters entities refer to phonetic characters The current GRAALAN syllabification implementation for Romanian language contains 723 normal syllabification rules and the same number of phonetic syllabification rules VII THEINFLECTION RULES GRAALAN collection of inflection rules for a certain natural language contains the rules used to obtain all the Lexicon Inflection rules Compound rule inflection situations of the lemmas from the lexicon and also the associated inflected forms represented in normal alphabet and phonetic alphabet The inflection rules are organised on four levels compound rules basic rules elementary rules and variants see Fig 4 Each le
29. ss masculine feminine neuter for gender plural singular for number etc organised under the form of an AVT 6 where the attributes correspond to the morphological categories and the attribute values correspond to the morphological attribute values An AVT has as nodes attributes and attribute values We can define very simplified the structure of an AVT in BNF notation 1 lt AVT gt gt lt attribute gt lt attribute value list gt 2 lt attribute value list gt gt lt attribute value element gt lt attribute value list gt lt attribute value element gt 3 lt attribute value element gt gt lt attribute value gt lt AVT gt lt attribute value gt A path in an AVT is formed by a sequence of lt attribute gt lt attribute value gt pairs Such a sequence is named EC Exclusive Combination An EC represents an inflection situation GRAALAN Compiler GRAALAN Compiler GRAALAN Syllab rules Compiler XML LKB GRAALAN Compiler GRAALAN Compiler GRAALAN Compiler Figure 2 Inflection process in a complex NLP System Alphabet XML LKB Morph Config XML LKB Inflection rules XML LKB Lexicon XML LKB Inflection forms XML LKB The expanded paradigm of a word covers a number of inflection situations form the morphological configurator i e a number of ECs Therefore the morphological configurator represents the set of all the inflection situations that al
30. y inflection rule Inflection form reference Variant Figure 4 A fragment of a basic inflection rule for Romanian language is the following Basic Rule Subst_mascl clasa substantiv tip substantiv comun animatie animat inanimat gen masculin numar singular caz nominativ Inflection rules articulare nearticulat EtL1 alphabetic phonetic hotarat EtS11 if amp consonant alphabetic insert ul last letter is a consonant b iat elev phonetic insert amp close_back_rounded amp alveolar_lateral_approximant if 1 alphabetic insert ul last letter is i ochi phonetic insert amp close_back_rounded amp alveolar_lateral_approximant genitiv dativ acuzativ plural We give below some figures in order to have an idea on the complexity of inflection rules for Romanian language Inflection situations EC that have inflection rules 19 202 Inflection situations that do not have inflected rules 260 these situations are procedural treated for example some complex numerals e Variants 28 317 e Multiword variants 19 935 e Defective situations 2 936 e Mono word variants 8 382 e Multiword variants with 2 words 7 785 e Multiword variants with 3 words 6 554 e Multiword variants with 4 words 3 196 e Multiword variants with 5 words 1 908 e Multiword variants with 6 words 492 VIN THEINFLECTION FORMS The
Download Pdf Manuals
Related Search