Text Generation

Investors, before making an investment decision, rely on many sources of information, including the Web, which in current times has become an important mean of mass production and dissemination of information for the financial/stock market. In this work, we study techniques of Opinion Mining, applied to investment analysis, based on news sources disseminated on the Web, in order to achieve a better consumption and processing of this information, extracting the relevant part of the huge amount of unstructured data (text) generated on the Web.

Data imbalance is a frequently occurring problem in classification tasks where the number of samples in one category exceeds the amount in others. Quite often, the minority class data is of great importance representing concepts of interest and is often challenging to obtain in real-life scenarios and applications. Imagine a customers’ dataset for bank loans-majority of the instances belong to non-defaulter class, only a small number of customers would be labeled as defaulters, however, the performance accuracy is more important on defaulters labels than non-defaulter in such highly imbalance datasets. Lack of enough data samples across all the class labels results in data imbalance causing poor classification performance while training the model. Synthetic data generation and oversampling techniques such as SMOTE, AdaSyn can address this issue for statistical data, yet such methods suffer from overfitting and substantial noise. While such techniques have proved useful for synthetic...

Introduction: Traumatic brain injury (TBI) is a silent epidemic that affects at least 1.4 million people in the U.S. each year. An estimated 5.3 million Americans are living with disabilities resulting from TBI, and another 80,000 to 90,000 people join their numbers annually (Langlois et al, 2005). Research that provides a deeper understanding of the patient’s perspective can provide clinicians and others with valuable insights into the lives of TBI survivors and the

This paper reports a study on how ESL/EFL writers use their L1 (®rst language) when composing in their L2 (second language) and how such L1 use is affected by L2 pro®ciency and writing tasks. Sixteen Chinese EFL learners were asked to compose aloud on two tasks, narration and argumentation. Analyses of their think-aloud protocols revealed that these student writers had both their L1 and L2 at their disposal when composing in their L2. They were more likely to rely on L1 when they were managing their writing processes, generating and organizing ideas, but more likely to rely on L2 when undertaking taskexamining and text-generating activities. Additionally, more L1 use was found in the narrative writing task than in the argumentative writing. Finally, the think-aloud protocols re¯ected that L1 use decreased with the writer's L2 development, but the extent of the decline of L1 use in individual activities varied. Based on these ®ndings, an L2 composing process model is proposed. # Research on the L2 writing process has thrived since the early 1980s. The focus of attention has been, however, mainly on the similarities between L1 and L2 writing processes despite the``salient and important differences'' between them (Silva, 1993). One important difference between L1 and L2 writing processes is that L2 writers have more than one language at their disposal; that is, they may use both L1 and L2 for cognitive operations when they are composing in the L2. This difference has received limited attention from SLA : S 1 0 6 0 -3 7 4 3 ( 0 2 ) 0 0 0 8 4 -X researchers, resulting in little understanding of the unique features of L2 writing and a lack of a coherent, comprehensive L2 writing theory. The present study, therefore, focused on the use of L1 in the L2 composing processes of 16 advanced L2 writers, with a view to arriving at a better understanding of the distinct nature of L2 writing.

Taking the role of, in the words of Koneski, "our only complete homeland", the Macedonian language in the 21st century faces numerous existential challenges which, thanks to modern technologies or more precisely, artificial intelligence that would operate in Macedonian, we assume that they can at least be overcome partially, and this is the belief upon which this work is based. The sustainability, the ability to survive, and the overcoming of these challenges in the Macedonian literary language require a more serious and financially convenient institutional treatment that would allow researchers to work more broadly and an approach at the level that the Macedonian language and, consequently, Macedonian culture and civilization deserve.
The development of OAIML - Operational Artificial Intelligence in Macedonian Language, facilitates a technological step forward with potentials for application not only in relation to the Macedonian language, but also to the other South Slavic, Slavic, i.e. Indo-European and other less related languages. This kind of technology provides a mechanism for potentially unlimited writing or generating of texts from a wide range of areas, from journalistic articles, promotional texts and literary works to application in the context of generating texts for scientific papers, legal acts and legal solutions, economic analyzes, reports etc.
The common essential constituents of such a program are: automatic Natural-language generation (aNLG), automatic Natural language processing (aNLP) and various types of databases that can be exploited in 3 ways - generating, summarizing and/or editing various types of text. 
This paper briefly explains the constituents of such a program based on existing efforts in this area, and some aspects of existing programs are given and considered as a potential basis and an example for creating an equivalent program for the Macedonian language.

The Voynich manuscript is a medieval book written in an unknown script. This paper studies the relation between similarly spelled words in the Voynich manuscript. By means of a detailed analysis of similar spelled words it was possible to reveal the text generation method used for the Voynich manuscript.

Ghosts, Robots, Automatic Writing: An AI Study Level Guide by Anne Alexander, Caroline Bassett, Alan Blackwell, and Jo Lindsay Walton

"As far as academic ‘research outputs’ go, this is a strange one, really.
The result of a two-year long collaboration between the authors, shaped
by discussion with dozens of participants in an online workshop series
hosted by Cambridge Digital Humanities Learning Programme and
CRASSH, it sits somewhere between speculative fiction, scholarly essay
and a primer for that elusive being ‘the general public’ about an emerging technology. The actual date of publication is 2021, but readers are
asked to imagine when they read it that they are living in a future where
Automatic Writing has made it into school literature curricula and onto
the syllabus for undergraduate degrees.
In common with much writing about imagined futures, the essays in
this booklet highlight current debates and concerns by placing them in
a new context. By exploring tensions in the research communities which
have coalesced around the project of creating ‘Large Language Models’
through fiction, we also hope to encourage deeper and more critical
engagement with this field of research, particularly by researchers from
the Humanities." - Anne Alexander

The Expressive Intelligence Studio is developing a new approach to freeform conversational interaction in playable media that combines dialogue management, natural language generation (NLG), and natural language understanding. In this paper, we present our method for dialogue generation, which has been fully implemented in a game we are developing called Talk of the Town. Eschewing a traditional NLG pipeline, we take up a novel approach that combines human language expertise with computer generativity. Specifically, this method utilizes a tool that we have developed for authoring context-free grammars (CFGs) whose productions come packaged with explicit metadata. Instead of terminally expanding top-level symbols—the conventional way of generating from a CFG—we employ an unusual middle-out procedure that targets mid-level symbols and traverses the grammar by both forward chaining and backward chaining, expanding symbols conditionally by testing against the current game state. In this paper, we present our method, discuss a series of associated authoring patterns, and situate our approach against the few earlier projects in this area.

In this project, we have developed a sign language tutor that lets users learn isolated signs by watching recorded videos and by trying the same signs. The system records the user's video and analyses it. If the sign is recognized, both verbal and animated feedback is given to the user. The system is able to recognize complex signs that involve both hand gestures and head movements and expressions. Our performance tests yield a 99% recognition rate on signs involving only manual gestures and 85% recognition rate on signs that involve both manual and non manual components, such as head movement and facial expressions.

Interactive fiction and adventure video games are narrative genres which provide the player with the option of acting as the main character of the story. However these genres do not fully match the expectations of their authors and readers because the player cannot deeply affect the storyline.

We extend the Expressionist project, and thereby the re-emerging area of grammar-based text generation, by applying a technique from software verification to a critical search problem related to content generation from grammars. In Expressionist, authors attach tags (corresponding to pertinent meanings) to nonterminal symbols in a context-free grammar , which enables the targeted generation of content that expresses requested meanings (i.e., has the requested tags). While previous work has demonstrated methods for requesting content with a single required tag, requests for multiple tags yields a search task over domains that may realistically span quintillions or more elements. In this paper, we reduce Expressionist grammars to symbolic visibly pushdown automata, which allows us to locate in massive search spaces generable outputs that satisfy moderately complex criteria related to tags. While the satisficing of more complex tag criteria is still not feasible using this technique, we forecast a number of opportunities for future directions.

This is an important book. On the surface, it is a survey and summary of work related to a major computational linguistics effort--the Penman project at the University of Southern California. Beneath the specific title and the apparently narrow subject matter, however, lies a general point about the field of computational linguistics: many important aspects of language are not addressed by the generative tradition that has dominated the field. This book aims to demonstrate that other types of linguistic description are available to serve as the basis for a computational linguistic treatment of these issues.

This paper reports a study on how ESL/EFL writers use their L1 (®rst language) when composing in their L2 (second language) and how such L1 use is affected by L2 pro®ciency and writing tasks. Sixteen Chinese EFL learners were asked to compose aloud on two tasks, narration and argumentation. Analyses of their think-aloud protocols revealed that these student writers had both their L1 and L2 at their disposal when composing in their L2. They were more likely to rely on L1 when they were managing their writing processes, generating and organizing ideas, but more likely to rely on L2 when undertaking taskexamining and text-generating activities. Additionally, more L1 use was found in the narrative writing task than in the argumentative writing. Finally, the think-aloud protocols re¯ected that L1 use decreased with the writer's L2 development, but the extent of the decline of L1 use in individual activities varied. Based on these ®ndings, an L2 composing process model is proposed. # Research on the L2 writing process has thrived since the early 1980s. The focus of attention has been, however, mainly on the similarities between L1 and L2 writing processes despite the``salient and important differences'' between them (Silva, 1993). One important difference between L1 and L2 writing processes is that L2 writers have more than one language at their disposal; that is, they may use both L1 and L2 for cognitive operations when they are composing in the L2. This difference has received limited attention from SLA : S 1 0 6 0 -3 7 4 3 ( 0 2 ) 0 0 0 8 4 -X researchers, resulting in little understanding of the unique features of L2 writing and a lack of a coherent, comprehensive L2 writing theory. The present study, therefore, focused on the use of L1 in the L2 composing processes of 16 advanced L2 writers, with a view to arriving at a better understanding of the distinct nature of L2 writing.

This white paper is part of a series that promotes knowledge about language technology and its potential. The availability and use of language technology in Europe varies between languages. Consequently, the actions that are required to further support research and development of language technologies also differ. The required actions depend on many factors, such as the complexity of a given language and the size of its community.

META-NET, a Network of Excellence funded by the European Commission, has conducted an analysis of current language resources and technologies in this white paper series. The analysis focused on the 23 official European languages as well as other important national and regional languages in Europe. The results of this analysis suggest that there are tremendous deficits in technology support and significant research gaps for each language. The given detailed expert analysis and assessment of the current situation will help maximise the impact of additional research.

This paper reports a study on how ESL/EFL writers use their L1 (®rst language) when composing in their L2 (second language) and how such L1 use is affected by L2 pro®ciency and writing tasks. Sixteen Chinese EFL learners were asked to compose aloud on two tasks, narration and argumentation. Analyses of their think-aloud protocols revealed that these student writers had both their L1 and L2 at their disposal when composing in their L2. They were more likely to rely on L1 when they were managing their writing processes, generating and organizing ideas, but more likely to rely on L2 when undertaking taskexamining and text-generating activities. Additionally, more L1 use was found in the narrative writing task than in the argumentative writing. Finally, the think-aloud protocols re¯ected that L1 use decreased with the writer's L2 development, but the extent of the decline of L1 use in individual activities varied. Based on these ®ndings, an L2 composing process model is proposed. # Research on the L2 writing process has thrived since the early 1980s. The focus of attention has been, however, mainly on the similarities between L1 and L2 writing processes despite the``salient and important differences'' between them (Silva, 1993). One important difference between L1 and L2 writing processes is that L2 writers have more than one language at their disposal; that is, they may use both L1 and L2 for cognitive operations when they are composing in the L2. This difference has received limited attention from SLA : S 1 0 6 0 -3 7 4 3 ( 0 2 ) 0 0 0 8 4 -X researchers, resulting in little understanding of the unique features of L2 writing and a lack of a coherent, comprehensive L2 writing theory. The present study, therefore, focused on the use of L1 in the L2 composing processes of 16 advanced L2 writers, with a view to arriving at a better understanding of the distinct nature of L2 writing.

Qualitative simulation models can play a useful role in computer-based learning environments, since they explicitly represent the domain knowledge required for causal reasoning about system structure and behaviour. However, the amount of information in a simulation is often large, which makes it difficult to transform the computer's reasoning trace to effective explanations. Besides textual explanations, graphical representations play an important role in the communication of knowledge. For example, state-transition diagrams, causal networks, structural hierarchies, tables and graphs all use different visual primitives to create a context for communication, and to denote relationships between the different kinds of entities. Our goal is to develop generic mechanisms for automatic visualization of qualitative simulation models, and to integrate these into a framework for generating multimedia (text and graphics) explanations. Currently, we have developed an interactive model inspection tool, VisiGarp, which automatically generates several kinds of diagrams from a qualitative simulation model; a mouse and menu interface allows flexible switching between the different views on the model, providing both overview and detail. Further work includes layout optimization, aggregation and abstraction mechanisms, and integration of graphics and text generation.

In this paper, we describe issues in the translation of proper names from English to Chinese which we have faced in constructing a system for multilingual text generation supporting both languages. We introduce an algorithm for mapping from English names to Chinese characters based on (1) heuristics about relationships between English spelling and pronunciation, and (2) consistent relationships between English phonemes and Chinese characters.

In our current research into the design of cognitively well-motivated interfaces relying primarily on the display of graphical information, we have observed that graphical information alone does not provide sufficient support to users particularly when situations arise that do not simply conform to the users' expectations. This can occur due to too much information being requested, too little, information of the wrong kind, etc. To solve this problem, we are working towards the integration of natural language generation to augment the interaction functionalities of the interface. This is intended to support the generation of flexible natural language utterances which pinpoint possible problems with a user's request and which further go on to outline the user's most sensible courses of action away from the problem. In this paper, we describe our first prototype, where we combine the graphical and interaction planning capabilities of our graphical information system SIC! with the text generation capabilities of the Penman system. We illustrate the need for such a combined system, and also give examples of how a general natural language facility beneficially augments the user's ability to navigate a knowledge base graphically.

The objective of this study is to automatically extract annotated sign data from the broadcast news recordings for the hearing impaired. These recordings present an excellent source for automatically generating annotated data: In news for the hearing impaired, the speaker also signs with the hands as she talks. On top of this, there is also corresponding sliding text superimposed on the video. The video of the signer can be segmented via the help of either the speech or both the speech and the text, generating segmented, and annotated sign videos. We call this application as Signiary, and aim to use it as a sign dictionary where the users enter a word as text and retrieves sign videos of the related sign. This application can also be used to automatically create annotated sign databases that can be used for training recognizers.

Metamodeling is raising more and more interest in the field of language engineering. While this approach is now well understood for defining abstract syntaxes, formally defining concrete syntaxes with metamodels is still a challenge. Concrete syntaxes are traditionally expressed with rules, conforming to EBNF-like grammars, which can be processed by compiler compilers to generate parsers. Unfortunately, these generated parsers produce concrete syntax trees, leaving a gap with the abstract syntax defined by metamodels, and further ad-hoc hand-coding is required. In this paper we propose a new kind of specification for concrete syntaxes, which takes advantage of metamodels to generate fully operational tools (such as parsers or text generators). The principle is to map abstract syntaxes to concrete syntaxes via bidirectional mapping-models with support for both model-to-text, and textto-model transformations.

Ordering information is a difficult but important task for applications generating naturallanguage texts such as multi-document summarization, question answering, and conceptto-text generation. In multi-document summarization, information is selected from a set of source documents. However, improper ordering of information in a summary can confuse the reader and deteriorate the readability of the summary. Therefore, it is vital to properly order the information in multi-document summarization. We present a bottom-up approach to arrange sentences extracted for multi-document summarization. To capture the association and order of two textual segments (e.g. sentences), we define four criteria: chronology, topical-closeness, precedence, and succession. These criteria are integrated into a criterion by a supervised learning approach. We repeatedly concatenate two textual segments into one segment based on the criterion, until we obtain the overall segment with all sentences arranged. We evaluate the sentence orderings produced by the proposed method and numerous baselines using subjective gradings as well as automatic evaluation measures. We introduce the average continuity, an automatic evaluation measure of sentence ordering in a summary, and investigate its appropriateness for this task.

Resumen: Uno de los desafíos de la Generación de Lenguaje Natural es la adaptación de la estructura y las palabras de la salida lingüística a la habilidad del usuario, el contenido, el género apropiado, el estilo, etc. Nos centramos en la determinación de la estructura del discurso. En general, se supone que entre dos unidades de contenido ocurre siempre la misma relación de discurso. Propuestas que varían el tipo de relación discursiva y el orden de las proposiciones según la interpretación del contenido siguen siendo escasas. Sin embargo, tal interpretación es extremadamente importante especialmente si el contenido es altamente dinámico como por ejemplo, cuando los datos son series temporales. Presentamos un planificador de textos que considera las restricciones que imponen los datos dinámicos para tomar decisiones a cada etapa de la planificación, en particular para la selección de las relaciones discursivas y la ordenación de las proposiciones. Palabras clave: generación de textos, contenido dinámico, planificación RST, XSLT, información sobre la calidad del aire Abstract: One of Natural Language Generation's continuing challenges is to determine the structure and words of the generated linguistic output in accordance with the expertise of the user, the content, the appropriate genre, style, etc. We focus on the determination of the discourse structure. Most often, it is assumed that between two content units always the same discourse relation holds. Approaches in which the choice of discourse relations and the ordering of propositions depends on the interpretation of the content are still scarce. However, such an interpretation is extremely important especially if the content is highly dynamic as, e.g., in the case of data parameter time series. We present a text planner that takes into account the constraints imposed by dynamic data to make decisions at every stage of the text planning, and in particular, for the selection of discourse relations and the ordering of propositions.

In our current research into the design of cognitively well-motivated interfaces relying primarily on the display of graphical information, we have observed that graphical information alone does not provide sufficient support to users particularly when situations arise that do not simply conform to the users' expectations. This can occur due to too much information being requested, too little, information of the wrong kind, etc. To solve this problem, we are working towards the integration of natural language generation to augment the interaction functionalities of the interface. This is intended to support the generation of flexible natural language utterances which pinpoint possible problems with a user's request and which further go on to outline the user's most sensible courses of action away from the problem. In this paper, we describe our first prototype, where we combine the graphical and interaction planning capabilities of our graphical information system SIC! with the text generation capabilities of the Penman system. We illustrate the need for such a combined system, and also give examples of how a general natural language facility beneficially augments the user's ability to navigate a knowledge base graphically.

A novel random text generation model is introduced. Unlike in previous random text models, that mainly aim at producing a Zipfian distribution of word frequencies, our model also takes the properties of neighboring co-occurrence into account and introduces the notion of sentences in random text. After pointing out the deficiencies of related models, we provide a generation process that takes neither the Zipfian distribution on word frequencies nor the small-world structure of the neighboring co-occurrence graph as a constraint. Nevertheless, these distributions emerge in the process. The distributions obtained with the random generation model are compared to a sample of natural language data, showing high agreement also on word length and sentence length. This work proposes a plausible model for the emergence of large-scale characteristics of language without assuming a grammar or semantics.

Rhetorical Structure Theory (RST), a framework for analysing discourse structure above the clausal level, has been used extensively by both text analysts and researchers in text generation for the past decade. The proliferation of relations posited by various ...

In this article, some first elements of a computational modelling of the grammar of the Martiniquese French Creole dialect are presented. The sources of inspiration for the modelling is the functional description given by Damoiseau (1984), and Pinalie's & Bernabe's (1999) grammar manual. Based on earlier works in text generation (Vaillant, 1997), a unification grammar formalism, namely Tree Adjoining Grammars (TAG), and a modelling of lexical functional categories based on syntactic and semantic properties, are used to implement a grammar of Martiniquese Creole which is used in a prototype of text generation system. One of the main applications of the system could be its use as a tool software supporting the task of learning Creole as a second language. -- Nous présenterons dans cette communication les premiers travaux de modélisation informatique d'une grammaire de la langue créole martiniquaise, en nous inspirant des descriptions fonctionnelles de Damoiseau (1984) ainsi que du manuel de Pinalie & Bernabé (1999). Prenant appui sur des travaux antérieurs en génération de texte (Vaillant, 1997), nous utilisons un formalisme de grammaires d'unification, les grammaires d'adjonction d'arbres (TAG d'après l'acronyme anglais), ainsi qu'une modélisation de catégories lexicales fonctionnelles à base syntaxico-sémantique, pour mettre en oeuvre une grammaire du créole martiniquais utilisable dans une maquette de système de génération automatique. L'un des intérêts principaux de ce système pourrait être son utilisation comme logiciel outil pour l'aide à l'apprentissage du créole en tant que langue seconde.

In this article, we investigate the structural composition of the Voynich manuscript text, using statistical (network) analysis of word/token similarity. Our results support the so-called “hoax hypothesis,” i.e., interpretation of the text as a set of meaningless strings. Based on this analysis, we present a concrete text-generator algorithm (the “self-citation” process), easily executable without additional tools even by a medieval scribe. The analyzed text sample reproduces some of the statistical key properties of the Voynich manuscript; in particular, both of Zipf’s laws are fulfilled.

The paper discuses some problems of presentation and processing of linguistic knowledge, needed for the development of real-size Bulgarian linguistic resource to be used in a multilingual text generation system, covering software manuals sublanguage. The sublanguage volume is specified by corpus analysis. The text generation is based on the Systemic Functional Linguistics theory. A method for developing lexico-grammar resources by re-using an existing resource for another language is described and illustrated with three examples.

This diploma thesis explores subjectivity in the domain of so-called digital writing, that is, in texts of largely experimental nature generated by computer algorithms (or with their assistance). In order to do so, the thesis briefly covers the history of digital writing, its mediatic specificities, poetics as well as various theoretical and philosophical conceptualizations. Most importantly, it undertakes an analysis of a post-lyrical subject, a concept devised by Janez Strehovec, that is common to all cases of generative writing under focus. For its comparative analysis, the thesis deals with the recent works from contemporary creators who approach algorithmic textuality from variegated perspectives, incl. Nick Montfort, Allison Parish, Stephanie Strickland, Li Zilles, and Jörg Piringer. Texts generated by programs are conceived of as expressing a new, parasitic and prosthetic, genus of cyber-textual subjectivity that defies the traditional lyric and expands its pool “by other means,” as Marjorie Perloff would say. Such a tendency results in conceptually as well as formally complex literary corpus “infected” by – to further exploit the suggested metaphor – parasitic voices and prosthetic selves. Unlike in generic lyric, the post- lyrical subject surpasses the confines of poetry as genre; it is grounded in collectivity, multivocality, mutability, uncreativity and serves not only as a representation of the multifaceted – not to say cyborgian à la N. Katherine Hayles and Donna J. Haraway – postmodern psyche with respect to the form and message, but also, on the other hand, the means of production and chain of command behind its textual architecture. In the world of accelerated automation and augmentation, user interfaces, neural networks, machine learning and ubiquitous AI, this thesis aims to facilitate the contextualization of subjectivity in the realm of hybrid and generated texts and perhaps also to investigate subjectivity through the gaze of the mechanical Other.

A general, reusable computational resource has been developed within the Penman text generation project for organizing domain knowledge appropriately for linguistic realization. This resource, called the upper model, provides a domain-and task-independent classification system' that supports sophisticated natural language processing while significantly simplifying the interface between domain-specific knowledge and general linguistic resources. This paper presents the results of our experiences in designing and using the upper model in a variety of applications over the past 5 years. In particular, we present our conclusions concerning the appropriate organization of an upper model, its domainindependence, and the types of interrelationships that need to be supported between upper model and grammar and semantics.

In advanced XML transformer tools, XSLT rules are generated automatically after relating simple source and target XML documents. In this paper, we generalize this approach for the design of model transformations: transformation rules are derived semi-automatically from an initial prototypical set of interrelated source and target models. These initial model pairs describe critical cases of the model transformation problem in a purely declarative way. The derived transformation rules can be refined later by adding further source-target model pairs. The main advantage of the approach is that transformation designers do not need to learn a new model transformation language, instead they only use the concepts of the source and target modeling languages.

Experiment 1 assessed the time and effort allocated to writing subprocesses while generating written and verbal protocols over 10 weekly writing sessions. Within a 40-min session, planning time consumed about 45% in the first 5 min, but stabilized at near 300!& thereafter. Generating text initially consumed 400!& of the writers' time, peaked at 50%midway, and then declined to its original level. The time spent revising and reviewing was negligible early in writing sessions, but increased substantially late in the sessions. The highest and lowest quality documents could be differentiated on the basis of the amount of time the writers devoted to revising and to the magnitude of their RTs in a secondary interference task. Writers showed consistent, distinctive patterns of transitional probabilities between writing subprocesses both within and across sessions, yielding quantitative representations of their writing styles. In Experiment 2, writers overestimated the amount of time they devote to revising and overestimated the amount of effort they allocate to planning and text generation. Their estimations did not improve after 10 weeks of composing. A time-and-effort-based analysis of writing is proposed to account for these data.

In this project, we have developed a sign language tutor that lets users learn isolated signs by watching recorded videos and by trying the same signs. The system records the user's video and analyses it. If the sign is recognized, both verbal and animated feedback is given to the user. The system is able to recognize complex signs that involve both hand gestures and head movements and expressions. Our performance tests yield a 99% recognition rate on signs involving only manual gestures and 85% recognition rate on signs that involve both manual and non manual components, such as head movement and facial expressions.

this paper on the development and evolution of such a component, the surge system. surge has been widely distributed in the generation community and has been embedded into several complete systems. The goals of reusability and wide coverage have led to a large system and many of the issues faced during the development of the system are issues common to the development of large software systems. From the linguistic side, they have also led to the definition of an input specification language of a relatively low level of abstraction and to a pragmatic approach c fl 1999 Association for Computational Linguistics Computational Linguistics Volume 99, Number 4 to grammar development that integrates several distinct theories into one computational engine. Indeed, while overall surge is organized around the lines of a systemic functional grammar, we have found it efficient to borrow from other linguistic schools (HPSG (Pollard and Sag, 1994)) and descriptive works (Quirk et al., 1985) to im...

Data imbalance is a frequently occurring problem in classification tasks where the number of samples in one category exceeds the amount in others. Quite often, the minority class data is of great importance representing concepts of interest and is often challenging to obtain in real-life scenarios and applications. Imagine a customers’ dataset for bank loans-majority of the instances belong to non-defaulter class, only a small number of customers would be labeled as defaulters, however, the performance accuracy is more important on defaulters labels than non-defaulter in such highly imbalance datasets. Lack of enough data samples across all the class labels results in data imbalance causing poor classification performance while training the model. Synthetic data generation and oversampling techniques such as SMOTE, AdaSyn can address this issue for statistical data, yet such methods suffer from overfitting and substantial noise. While such techniques have proved useful for synthetic...

We present a new approach to extracting information from unstructured documents based on an application ontology that describes a domain of interest. Starting with such an ontology, we formulate rules to extract constants and context keywords from unstructured documents. For each unstructured document of interest, we extract its constants and keywords and apply a recognizer to organize extracted constants as attribute values of tuples in a generated database schema. To make our approach general, we fix all the processes and change only the ontological description for a different application domain. In experiments we conducted on two different types of unstructured documents taken from the Web, our approach attained recall ratios in the 80% and 90% range and precision ratios near 98%.

In this paper, we present the first sizable grammar built for Vietnamese using LTAG, developed over the past two years, named vnLTAG. This grammar aims at modelling written language and is general enough to be both application- and domain-independent. It can be used for the morpho-syntactic tagging and syntactic parsing of Vietnamese texts, as well as text generation. We then