The Fourth International Conference on Future Generation Communication Technologies (FGCT 2015)
Current Communication Technologies in Language Processing Jolanta Mizera-Pietraszko, Institute of Mathematics and Computer Science, Opole University, Opole, Poland http://www.math.uni.opole.pl/~jmizera/ Email:
[email protected] Communication technologies rely on language processing, mainly text that is transformed into speech automatically, annotations of image, or graphical objects. Video is a specific multimedia form of information in which voice is recorded simultaneously with the vision of the objects being in motion. Consequently, the quality of the communication system output is determined by the numerous tools for language processing.
Abstract — Even the most cutting-edge communication-mediated technology like satellite navigation for orbit positioning, pedestrian movement recognition systems based on inertial sensors, 5G systems, let alone medical devices for coordination of human organs functionality would not be invented without technologies for language processing as an information source between humans and communication systems. Regardless of the way we communicate that is via emails, website short tweets, video conferencing systems, social networking, blogs, instant messaging through websites or mobile applications, or texting only, we use a language that is processed by computer system. Thus, the keynote paper discusses language processing technologies as the core of any other communication between human beings or human-machine. From the user’s perspective communication technologies simply integrate computer science and linguistics, but that is a very broad scope of knowledge. Technologies of a particular language rely on computational linguistics, natural language processing including also speech processing, information retrieval, knowledge acquisition and machine translation tools.
For the purpose of linguistic analysis defined are the levels representing particular language aspects such as morphology, semantics and syntax. Language as such is a system consisting of the elements like words, the rules of these words order called semantics or contextual meaning, grammar rules called syntax and finally merging the words into semantic groups making use of the grammar rules for sentence building. Having the levels of language analysis defined, we may correlate them with the commonly exploited tools responsible for processing this language: 1 - stemmer – at the morphology level is a tool for reducing the word forms to their stems in order to facilitate the analysis of the word class properties and the range of its meanings,
Keywords— Computer-Mediated Communication, Language Processing, Intelligent Information Systems, Computational Linguistics, Machine Translation, Information Retrieval
I.
2 – at the semantic level three tools are utilized together:
INTRODUCTION
a)
token as an elementary unit represented by a character or a word. In text normalization process called linguistic preprocessing, text is decomposed into tokens while diacritics, capital letters, accents and abbreviations are removed b) lexer for converting sequence of characters into the sequence of tokens quite useful also for a lexical analysis c) parser that is an extra tool for semantic analysis, however, when added to token and lexer, it supports syntactic analysis as well. In such a case, the word sequence determines generation of a sentence structure whose elements are part of speech. With regards to machine translation process, in the next phase, the sentence structure that has just been generated, is transposed onto the syntactic structure of the sentence in the target language. Therefore, the joint utilization of these tools grants the conceptual knowledge about the structure type, and in a broader
The potential of communication technology nowadays relies mainly on computers’ performance capacity especially those with the technical specification of processing natural language. Nonetheless, not only technologies, but especially the language tools determine the communication system efficiency resulting in the popularity of the particular language in the net. On the other hand, 5th generation wireless systems constitute the most influential trend in development of information and communication technology infrastructure that accounts for a high data transmission speed of more than a hundred MB/s to support almost million simultaneous connections worldwide, flexible protocols, advanced Physical Layers (PHY) and MAC technology, innovative security methodologies, intelligent terminals [6]. Therefore, communication intensively develops in two directions simultaneously: linguistic technologies usually support expansion of electronic devices infrastructure improving our life styles substantially.
978-1-4799-8267-7/15/$31.00 ©2015 IEEE
79
spectrum, about the target language, let alone the language pair phenomena. 3 – wrapper is a specific tool at the syntax level to faciliate data extraction from the predefined information sources making use of the collocations being a word type order 4 – merger – this tool supports the process of some simple sentences being merged into a complex sentence.
cohesion. The words being the nodes of the functional structure are in bold to emphasize their role in the example sentence. Each word is related to a function numbered from 1 to 6. For example, Pron form: This can be explained as a determiner This proceeding infinitive be that is predefined by a lexical function PRED, which is in such a case Present Tense, 3rd Person Singular tense: pres, num:sq, pers:3 The level of the tree structure represents a position of the word in the sentence. Syntax Tree is created in a form of a sentence net, whose nodes are the words while the net arrangement determines the sentence structure that is This to That. The arrangement is sequential because the sentences are simple. However, the remaining part is a complex sentence conjunct with and that has an equipotent position in the sentence structure. Such a process is called sentence parsing. Pragmatically, language processing tools determine quality of the system output so that they are applicable to one language only. Even though the tools vary from one to another like for instance in case of a word analysis Porter stemmer, which transforms the word into analysi (sis si), from Lovins stemmer that the same word is transformed into analys (sis s) due to their different rules. Still more and more innovative tools constitute a dominating factor of the processing efficiency. More popular languages have the greatest number of tools that are of the same kind since the competitiveness ensures the quality. Thus, our motivation is to indicate and analyse the broad spectrum of present areas of applications of the future communication technologies in order to observe and envisage the trends worldwide. The paper is organized as follows: section II concerns communication-driven technologies, section III examines language-supported communication technologies. In section IV investigated is multilingual environment in which natural language is processed in some specific ways, quite different from monolingual environment. Section V deals with with present trends in online machine translation systems. Section VI moves to future technological advancements in ICT and finally section VII concludes the keynote. The paper is finalised in the section presenting a short BIO of the Author and her research profile.
This software that takes two strings of space separated words as input and aligns matching words between the two strings. pron_form : this pred : be tense : pres num : sg pers : 3 xcomp : subj : pred : pro pron_form : this pred : software num : sg pers : 3 relmod : topicrel : pred : pro pron_form : that coord : 1 : subj : _6580 pred : take tense : pres num : sg pers : 3 obj : spec : quant : pred : two pred : string num : pl pers : 3 adjunct : 2 : pform : of obj : adjunct : 3 : pred : space num : sg pers : 3 4 : pred : separate tense : past pred : word num : pl pers : 3 adjunct : 5 : pform : as obj : pred : input num : sg pers : 3 6 : subj : _6580 pred : aligns tense : pres num : sg pers : 3 obj : adjunct : 7 : pred : matching 8 : pform : between obj : spec : det : pred : the quant : pred : two pred : string num : pl pers : 3 pred : word num : pl pers : 3 subj : _6580 coord_form : and resolved : topicrel
II.
COMMUNICATION-DRIVEN TECHNOLOGIES
In this section analyzed are current research directions in communication technologies not supported linguistically, but those which supplement some disorders with new solutions. Patients with cerebral palsy can supplement their spoken communication with augmentative and alternative communication systems based on symbols and phrase suggestions [2]. Depre SD (Depression Syndrome Detection) is a system aimed at helping people who suffer from some emotional problems to prevent depression. Internet forum users and micro bloggers express their feelings freely whereas they are unable to share their problems with others in direct conversation. Depre SD analyses their posts by breaking the text into chunks which represent the words of sadness.
Fig. 1. Syntax Tree of the source sentence and its functional structure
Lexical Functional Grammar [5] provides an explanation for processing language from the perspective of positioning text units according to the function of it in a phrase, expression or a sentence. Fig. 1 shows an example sentence syntax structured whereas the last part provides the role of each text grammatical unit in the sentence on the whole called
80
Frequencies of some words indicate symptoms of depression [3]. Regarding speech recognition, energy and speed of utterances is analyzed based on some datasets in several languages to prove Gutenberg-Richter law that is a robust of speech synthesis technologies by providing hints to humanize automatic speech [4]. Blood cancer cells are recognized in image screening for analyzing morphological features like large area of cytoplasm and a small area of lymphocytes, circumference of cell compartment, shape, rectangularity of a cell compartment, elongation, a ratio of nucleus and cytoplasm. To improve the methodology of cancer identification, some chromatic features are added such as darker than normal nuclear colour of a lesion cell [7]. Secure DV-Hop localization scheme for fighting wormhole attacks in wireless sensor networks is an example of effective mechanism to support communication systems [8]. Chats turn out to be an effective medium for therapy of the patients with bulimic symptomatology whose transcripts are analyzed with multiple linear regression function for words related to family origin as a prediction symptom of the eating disorders improvement [9]. Listing devices exposed on sensitive data theft especially ebank transactions, sometimes offered via the eGovernment services; users, transparent encryption of outbound data or predefining some sensitive data types are the current technologies aimed at prevention of data leakage [10]. Communication system Radiologue at San Francisco General Hospital keeps in still contact, in particular for alerts, Radiology staff with Emergency Department physicians and nurses. ODBC generic connection links the two departments by supporting vendor-specific interfaces [11]. In Henan Polytechnic University, China Bipolar chaotic pulse position modulation communication system has been developed. The system deploys cyclic low-density paritycheck code (LDPC) to mitigate the noise and thus minimizing the channel distortion, if any [12]. In communication-driven systems it is always the information that plays the key role in the areas of their possible applications. Although, the language itself is not linguistically analyzed, some grammar (e.g. some word semantics) is processed making the systems nationally oriented. III.
the PubMed database show that Gene Mutations and Phenotypes, Organisms and their Pathogenicity, Gene Products and their Roles, F-mean is more than 0.9 for genes and operons, while 0.75 for gene products [14]. Machine learning-supported NLP algorithm indicates an improve in tromboembolic disease diagnosis from radiology reports based on the analysis of relations between the concepts. For identification of pulmonary embolism the Fmean reaches almost 1, another 1 for deep vein thrombosis, and 0.8 for incidental clinically relevant findings [15]. Data structures are processed from clinical text to standard text for the further usage of generating both structured and unstructured data in variety of settings [16]. Word sense disambiguation is a subject of analysis of big data from a parallel corpus achieving promising results in granularity of semantic processing [17]. This technology indicates a trend towards increasing popularity of translingual corpora processing. Unlike communication-driven technologies, the studies presented in this paper prove that efficiency of language processing requires integration of information extraction with machine learning, information retrieval and more and more often of translingual web documents processing. IV.
NATURAL LANGUAGE PROCESSING IN MULTILINGUAL ENVIRONMENT
Multilingual enviroment affects natural language processing predominantly. Perhaps for this reason, soon before 2000 a new science called Cross-Language and Multilingual Information Access, also Translingual Engineering was developed. Processing machine translation is grounded on statistical models SMT, which generally refers to machine learning that follows the rule; the more linguistic structures of an indentified syntax is entered to the system, the higher translation quality is automatically produced. That is one of the factors generating translation quality; however it cannot be regarded separately from the others [18]. Another factor is a selection of search modes – as a rule, the user prefers single, but monosemic keywords to express his or her information need. Logical operators invariably inbuilt in searching algorithms have tendency to disturb matching the keywords with the equivalent in the posting list of the index [19]. Figure 2 shows a diagram factors that determine quality of Machine Translation and consequently affects multilingual information retrieval. Despite the decreasing number of online MT systems in which a number of words in a source language is limited, this factor impedes language processing. Producers of AltaVista BabelFish inform their users „Use correct spelling, grammar, and punctuation for the highest quality translations. Additionally, avoid slang and use short, clear sentences.” while Google reserves a maximal number of the keywords, especially when complex sentences are entered [20].
LANGUAGE-SUPPORTED COMMUNICATION TECHNOLOGIES
Most communication systems however, are languagesupported, so that their efficiency is in strict relation to the quality of linguistic analysis which constitutes its potential. Thus, this section provides some examples of such technologies in order to underline the difference from the previously described communication-driven technologies. Theory of inventive problem solving serves for the purpose of estimating ideality of patent description to stimulate the inventors creativity. The solution proposed integrates NLP of the patent description with machine learning and neural networks [13]. Nonetheless, even much more complex is a study on an enteropathogenic bacteria Escherichia coli and Salmonella spp. Information extracted from the abstracts of
81
similarity of the text passage indexed in posting list l with the keywords to the input query in the target language. A position of the keywords in the he query and the answer determines the system’s confidence score and consequently the position in the ranking list. Lin&Katz algorithm [21] is one of the commonly applied appl for document classification of QA systems: (1) where is system’s confidence score, A - a set of keywords, N - a number of the words in the text, while is a number of f co-occurrences of a word in the text. Fig. 2. Diagram of factors that determine quality of Machine Machin Translation
In 2002, MT (Machine Translation) system Systran [1] as one of the first, incorporated following tools for processing pair phenomena only: language-pair 1 –language recognizer for automatic detection of a source language 2 – spell checker which relies on tolerant retrival search mode where asterix replaces all the possiblities p 3 – word deliminer – for identification of words as units in the languages whose grammar rules determine the character sequence 4 – lemmatizer as a tool for word detection and for creating the word inflectional forms 5 – text synthetizer is used for creating word forms in a target language 6 – semantic domain recoognizer – this tool identifies a domain of the text on the basis of the language technical terms pair phenomena is a set of linguistic features Language-pair related directly to similarities and differences between be the particular languages L1 as a source language and L2 as a target language.
An illustrative example of quite complex language processing in multilingual environment is question-answering question system whose typical architecture is presented in Figure F 3. Translation model, usually hybrid, as the first phase pha quarantees the baseline at the output for further processing. Index Building
Retrieval Model Question
Machine Translation
Question Analysis Answer Snippet
Web
Document Collection
Answer Validation
Answer Extraction
Fig. 3. Architecture of a typical Multilingual Question-Answering Answering System
The remaining QA components are processed contingened contingen upon the technologies chnologies described in Table I. The cells in dark indicate that the two last components Answer Extraction Extrac and Validation are not preceded by Question Analysis nor Linguistic indexing process, since they are carried out manually using Question Taxonomy. Next, a parser identifies id a sequence of tokens in the view of syntax to match the query to the parts-of-speech.
Modularity of the QA system architecture constitutes its simplicity in expanding, updating or exchanging some som of the components. The components omponents in bold are related to bilingual processing as a standard. Adding POS Tagger before translation and after the Question Analysis components improves the remaining parts of the processing. Question analysis is founded on the taxonomy of question classes with the corresponding well-defined defined notions as in Figure 4. The notion is two-fold; fold; the question class and the syntax connected with this selective class. Question class
TABLE I.
Notion
Definition – What is ...?
Definition – (Object) is …
Person - Who..?
Name – (Person)…
Reason – Why…?
Purpose – Because…
Location – Where…?
Region – (Region name) is…
Factoid – Does/Do…?
Affirmation/Negation –Yes/No Yes/No
Question analysis
Linguistic indexing
Answer extraction
MaxEnt Classification Automatic generation Statistical analysis Parsing
Words
Fragmentation
Lemmatization
N-grams
Fig 4. Question Taxonomy as a component of Question--Answering System
Retrieval model for QA is often a shallow technique in which documents are classified according to syntactic syntact
Lemmatization
82
TECHNOLOGIES OF NLP IN QA
Lexical Morphem
Named Entity Recognition Predefined expressions Numeric expressions
Answer validation Machine learning Integrated classification Training MER Overload of information collection Lexical similarity
TABLE II.
Relational analysis Parts of speech
Manual patterns
Syntactic transformation Semantic parser
IDENTIFICATION TECHNICS OF ENGLISH MODEL OF GRAMMAR STRUCTURES EXTRACTED FROM THE EUROPARL COLLECTION WITH KWIC (KEYWORDS IN CONTEXT) METHOD COMPRISING LANGUAGE SUBCLASSES OF THE MORPHOLOGY, SEMANTICS, SYNTAX AND CONJUNCTION CLASSES
Syntactic similarity Semantic analysis Proof of theory Logic representation
SUBCLASS Acronyms Cleft sentences Morphemes Hyponymy Fixed phrases Collocations
MaxEnd defines maximum probability of the relevant response R to the query Q as in [22] where
P(R Q) we get
! "
!
#
(3)
where
is a model of searching the response and is an algorithm of R filtering. Linguistic indexing selects a language unit like a word, lexical or inflexion morpheme (lemmatization). Extraction of the text portions with the response (snippet) to the user’s question from multilingual document collection is processed according to categorization of the text in natural language (fragmentation) into unigrams or n-grams. Preferred extraction technique seems proper names, numerical expressions, or some fields identification (named entities recognition).
VI.
784, 1255, 4258, 9934 1732, 285, 222,1383, 50, 921, 222, 65 739 16156 52441, 121
FUTURE TECHNOLOGICAL ADVANCEMENTS IN ICT
Communication technologies are developing even more intensively than the users can adapt to their lifestyle, that is a few-year laptop is old enough to cause problems with the operating system updates, or compatibility with the new applications, let alone some unique devices. Astonishingly, some of them emerge as quickly as they disappear on the market, due to the lack of the customers’ interest – to name some of them, miniaturization of smartphones or PDAs, letter recognizers or transcribers. MMS seems not quite popular. Synchronization is a trend towards connecting users’ devices with their laptops whenever they are – TelNet is an example. Social networks and social media make millions of users keep in touch with the network community. Despite its limitations, cloud computing is actually a potential in business. Also GPS network supports many mobile applications nowadays. The most innovative communication technologies are: web intelligence empowers WWW with still newer IT products and services intelligent systems that reason and learn intelligently neural computing sending messages as if from one brain neuron to another granular computation for processing complex information by its granularity computational semiotics integrating linguistic levels with signs for improving communication quantum computing using transistors for encoding messages
Sentence parsing enables to identify a question subject, thus semantic and syntactic analysis is performed. Answer validation relies on semantic and syntactic similarity between the candidate answers. Machine learning is executed by collecting the answers and then selecting those which fulfil the criteria predefined. Processing collection with minimizing error rate is called MER training of the system. Lexical, semantic or syntactic similarities are the criteria of grammar rules between the query and target text. Overload of information is a technology of the system selection of the text portion with the greatest number of expressions in the query and the text. Proof of theory is applied both for extraction and validation as the system transforms the documents into their logic forms, then based on mathematical logics and the representation of the text portion, it proves which of the questions are representative [23]. V.
NUMBER OF OCCURENCES 755, 767, 69, 51, 73
For unidentifiable syntactically subclasses like idioms or simple sentence we intuitively extracted from the EUROPARL collection a representative sample of linguistic units with characteristics of the individual subclass. Comparative techniques with the same identifier of at least two subclasses are not equipotential with the same frequency of the category occurrences in the subclass because it is related to another linguistic unit. For instance, and is both in subclass conjunction and fixed phrase. Number of occurrences is the same 16,156. On the contrary to it, the extraction technique with the, which generates subclasses lexical cohesion, collocations and anaphora, shown different number of occurrences for each case, that is 4621, 45575 and 52441 respectively. Such a result indicates mapping of the subclasses sets.
P(R Q)=P(R T, RQS) (2) T is a question type while RQS, a remaining question
string. By maximizing probability of
SYNTACTIC IDETIFICATION Full stop + letter, EU, US, FR,UK It is, who, which, that -ed, -ble, -ly, -al, -ory, tion, -ity, -less Colon and the + two words, -y the
TECHNOLOGY OF SYNTACTIC IDENTIFICATION OF ENGLIFH LANGUAGE SYSTEM
We developed a holistic model of the English language system that comprises the following features: 1 – monothetic classification 2 – complete classification 3 – inductive inference 4 – deterministic inference 5 – Lexical Functional Grammar for language modeling Table II presents all the techniques used for extraction of each linguistic unit of the English language system.
83
[7]
amorphous computing with parallel processors facilitating messaging between the devices via links VII.
CONCLUSION
The approach presented in this paper relies mainly on analysis of the trends in information technologies, information and communication technologies and finally communication technologies with the aim at highlighting the standard algorithms, tools and processing paradigms that foster the potential of the output quality, especially when regarding multilingual environment. Presented areas of research applications indicate the trends in communication development to be envisaged for the future.
[8]
[9]
[10]
The contribution of this work to indicate that whatever one calls the technology that is IT, ICT, or CT, it is always the language which supports every other communication. Translingual Web accounts for surmounting language boundaries as the predominant trend in communication. Intensification of striving towards innovative solutions makes the users seem unable to follow the new trends, which results in slower sale by the producers than expected - this somehow explains why some technologies tend to disappear. VIII.
[11]
[12]
Y. Li, L. Zhang, H. Lu, Y. Kitazono, S. Yang1, S. Nakashima, and S. Serikawa, " A New Type of Using Morphology Methods to Detect Blood Cancer Cells, In J. Luo (Ed.):Soft Computing in Information Communication Technology, AISC 158, Springer-Verlag, pp. 17–25, 2012 T. Zhang, J. He, and Y. Zhang, " Secure DV-Hop Localization against Wormhole Attacks in Wireless Sensor Networks', n J. Luo (Ed.):Soft Computing in Information Communication Technology, AISC 158, Springer-Verlag, pp. 17–25, 2012 A. Mezei, H. Gulec, E. Czegledi, "Linguistic characteristics of patients with bulimic symptomatology in an online post-treatment program: an exploratory study", EATING AND WEIGHT DISORDERS-STUDIES ON ANOREXIA BULIMIA AND OBESITY Volume: 20 Issue: 1 pp. 63-70, 2015 A. Palazov, "Some Technologies for Information Security Protection in Weak-Controlled Computer Systems and Their Applicability for eGovernment Services Users", In: J. Camenisch, V. Kisimov, and M. Dubovitskaya (Eds.): iNetSec 2010, LNCS 6555, pp. 117–122, 2011, IFIP International Federation for Information Processing 2011 A. V. Rybkin, M. Wilson, "A Web-Based Flexible Communication System in Radiology", Journal of Digital Imaging, Springer, vol. 24, 2011, pp.890-896 H. Li, H. Liu, S. Vafi, "Bipolar chaotic pulse position modulation communication system based on cyclic LDPC", EURASIP Journal on Wireless Communications and Networking, vol.105, 2014
[13] Ch. Adams and D. Tate, " Computer-Aided TRIZ Ideality and Level of Invention Estimation Using Natural Language Processing and Machine Learning", In: © IFIP International Federation for Information Processing 2009, R. Tan, G. Cao, and N. León (Eds.): CAI 2009, IFIP AICT 304, pp. 27–37, 2009 [14] S. Zaremba, M. Ramos-Santacruz, T. Hampton, P. Shetty, J. Fedorko, J. Whitmore, J. M Greene, N. T Perna, J. D Glasner, G. Plunkett III, M. Shaker, D. Pot, " Text-mining of PubMed abstracts by natural language processing to create a public knowledge base on molecular mechanisms of bacterial enteropathogens", BMC Bioinformatics, Springer, vol. 10, pp.107, 2009 [15] A-D. Pham, A. Névéo, T. Lavergne, D. Yasunaga, O. Clément, G. Meyer, R. Morello, A. Burgun, " Natural language processing of radiology reports for the detection of thromboembolic diseases and clinically relevant incidental findings", BMC Bioinformatics, Springer, vol. 15, pp.266, 2014 [16] S. T Wu, V. C Kagga, D. Dligach, J J Masanz, P. Chen, L. Becker, W. W Chapman, G, K Savova, H. Liu, Ch. G Chute, " A common type system for clinical natural language processing", Journal of Biomedical Semantics, Springer, vol. 4, pp.1, 2013 [17] D. Tufi, "Word Senses: The Stepping Stones in Semantic-Based Natural Language Processing", In IFIP Intemational Federation for Information Processing, Volume 204, Artificial Intelligence Applications and Innovations eds. Maglogiannis I., Karpouzis K., Bramer M., (Boston: Springer), pp.575-582, 2006 [18] P.Koehn, "Statistical Machine Translation", Cambridge University Press, 2010 [19] M. Agosti, "Web Retrieval. Clues on IR systems and service architecture", European Summer School on Information Retrieval, Aussois, France, 2003 [20] J. Candela, "Machine Learning Challenges" Workshop, LNAI, SpringerVerlag, vol. 3944, 2006 [21] J. Lin, B. Katz, "Question Answering from the Web Using Knowledge Mining Techniques", Proceedings of the 12th International Conference of Information and Knowledge Management, 2003 [22] M. Heie., R. Nowak, D. Whittaker, S. Furui S., "Question Answering Experiments at Tokyo Institute of Technology", Working Notes for CLEF 2009, Cross-Language Evaluation Forum, TrebleCLEF Coordination Action under the Seventh Framework Programme of the European Commission, 2009 [23] I. Glockner, B. Pelzer, " Extending a Logic-Based Question Answering System for Administrative Texts:, in: Multilingual Information Access Evaluation Experiments 1., by Carol Peters, Di Nunzio M., Kurimo M., Mandl, LNCS vol. 6241, pp. 265-272. Heidelberg: Springer, 2010
ABOUT THE AUTHOR
Jolanta Mizera-Pietraszko holds a position of an Assistant Professor at the Faculty of Mathematics, Physics and Computer Science, Opole University, Poland. She invented a language and system-independent asymmetric translation technology called “Technology of Machine Translation Analysis with implementation of Language Pair Phenomena” (invention number P387576, registered and published by the Patent Office of the Republic of Poland). Her research interests are mainly focused on Information Technologies, Data Mining, Computational Linguistics, Natural Language Processing, Web Information Retrieval, Machine Translation, Artificial Intelligence, e-Learning, Electronic Textbooks, Multilingual Search Technologies, Parallel Languages, Bi-text processing, Multilingual Question-Answering Systems, and Multilingual Digital Libraries. REFERENCES [1] M Flanagan., S McClure "Systran and the Reinvention of MT", IDC Bulletin 6459, 2002 [2] S. C. Eduardo, G. Ramirez, A. Rafael, C. M. Jose; et al. "An augmentative and alternative communication tool for children and adolescents with cerebral palsy", BEHAVIOUR & INFORMATION TECHNOLOGY Volume: 34 Issue: 6 Special Issue: SI Pages: 632645, JUN 3 2015 [3] K. Christian; H. Robert, T. Wetter, "Screening Internet forum participants for depression symptoms by assembling and enhancing multiple NLP methods", COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE Volume: 120 Issue: 1 Pages: 27-36, JUN 2015 [4] L. Jordi, L. Bartolo, L. Lucas, "Scaling and universality in the human voice", Volume: 12 Issue: 105, APR 6 2015 [5] R. M Kaplan., J. Bresnan, "Lexical-Functional Grammar: A Formal System for Grammatical Representation", In Bresnan 1982b, 173–281, pp. 29–135, 1995 [6] C_I. Badoi, N.Prasad, V.Croitoru, R.Prasad, "5G based on cognitive radio", Wireless Personal Communication, April 2011, Volume 57, Issue 3, pp 441-464, 2010
84