Báo cáo khoa học: " a Tool for Teaching by Viewing Computational Linguistics"

Chia sẻ: Hongphan_1 Hongphan_1 | Ngày: | Loại File: PDF | Số trang:4

Thêm vào BST

Báo xấu

63
lượt xem 4
download

Download Vui lòng tải xuống để xem tài liệu đầy đủ

ProLiV - Animated Process-modeler of Complex (Computational) Linguistic Methods and Theories - is a fully modular, ﬂexible, XML-based stand-alone Java application, used for computer-assisted learning in Natural Language Processing (NLP) or Computational Linguistics (CL). Having a ﬂexible and extendible architecture, the system presents the students, by means of text, of visual elements (such as pictures and animations) and of interactive parameter set-up, the following topics: Latent Semantics Analysis (LSA), (computational) lexicons, question modeling, Hidden-Markov-Models (HMM), and Topic-Focus. These topics are addressed to ﬁrst-year students in computer science and/or linguistics. ...

Chủ đề:

Bình luận(0) Đăng nhập để gửi bình luận!

Lưu

Nội dung Text: Báo cáo khoa học: " a Tool for Teaching by Viewing Computational Linguistics"

ProLiV - a Tool for Teaching by Viewing Computational Linguistics Monica Gavrila Cristina Vertan Hamburg University, NATS Hamburg University, NATS Vogt-K¨ lln Str 30, 20251, Germany o Vogt-K¨ lln Str 30, 20251, Germany o gavrila@informatik. vertan@informatik. uni-hamburg.de uni-hamburg.de Abstract • the Latent Semantic Analysis (LSA) module and the computational lexicons module - for ProLiV - Animated Process-modeler of linguists, Complex (Computational) Linguistic Methods and Theories - is a fully modular, • the question modeling module - for computer ﬂexible, XML-based stand-alone Java scientists, application, used for computer-assisted • the Hidden-Markov-Models (HMM) module learning in Natural Language Processing and Topic-Focus module - for both computer (NLP) or Computational Linguistics (CL). scientists and linguists. Having a ﬂexible and extendible architec- ture, the system presents the students, by 2 The Learning Path means of text, of visual elements (such as pictures and animations) and of interactive For each module, the learning path is guided by parameter set-up, the following topics: lessons, a terminology dictionary and interactive Latent Semantics Analysis (LSA), (com- activities. Exercises and small tests can also be putational) lexicons, question modeling, integrated. Hidden-Markov-Models (HMM), and The lessons include text, pictures and ani- Topic-Focus. These topics are addressed mations. Hyperlinks between lessons ensure a to ﬁrst-year students in computer science concept-oriented navigation through the learning and/or linguistics. content. Additionally key terms within the content are linked with dictionary entries. 1 Introduction Three central issues guided the development of the ProLiV software: The role of multimedia in teaching Natural Language Processing (NLP) is demonstrated 1. choosing the most adequate means (text / pic- by constant development of software packages ture / animation) to represent lessons content, such as GATE (http://gate.ac.uk) and NLTK (http://nltk.sourceforge.net/ 2. designing the layout (quantity and size of index.html). Detailed information about vi- text, colors) in order to increase the learning sual tools for NLP, in particular about GATE, is success, to be found in (Gaizauskas et al, 2001). 3. in case of the animations, deﬁning its com- ProLiV is a Java application framework, devel- ponents and parameters (speed, animation oped in a three-year project (2005-2008) at the steps, and graphical elements) to maximize University of Hamburg. It helps ﬁrst-year stu- their impact on users. dents to understand and learn, in an easier man- ner, either complex linguistic theories used in NLP Regarding the second issue above-mentioned, (e.g. question modeling) or statistical approaches the layout of the modules follows part of the for computational linguistics (e.g. LSA, HMM). guidelines found in (Orr et al., 1994) and (Thi- The learning process is supported by modules bodeau, 1997). integrating text, visual and interactive elements. In Considering the current multimedia develop- its ﬁrst released version, ProLiV contains the fol- ment, the trend is using animations to improve the lowing modules: learning process. Animations are assumed to be 13 Proceedings of the ACL-IJCNLP 2009 Software Demonstrations, pages 13–16, Suntec, Singapore, 3 August 2009. c 2009 ACL and AFNLP
a promising educational tool, although their efﬁ- Representation Average Result ciency is not fully proved. Researchers, such as Deﬁnition (formula) 3.5 (Morrison, 2000), showed that animations can Picture 2.91 convey more information and be helpful when Animation 1 3.64 showing details in intermediate steps of a process, Animation 2 2.09 but when building an animation it is very impor- Animation 3 2.45 tant to consider the background of the student (e.g. linguistics, natural sciences) and his/her psycho- Table 1: Evaluation of the animations in the ma- logical functioning. The educational effectiveness trix multiplication (Animations 1 and 3 have no of the animations depends on how they interact user interaction; Animations 1 and 2 are more ab- with the learner. Depending on the student’s back- stract) ground, in order to have a helpful material, one has to carefully decide what information the ani- the learning process being adapted to the user’s mation contains. As our experiment showed (see rhythm. All the evaluation results can be seen in Section 2.1), depending on the student and his/her Table 1. In order to better see the inﬂuence of these background, an animation can improve the learn- representations in the learning process, statistical ing process, or bring nothing to it. We found no tests should be run. cases when the animation slowed down the learn- ing process. 3 System Architecture The system was experimentally used in semi- nars at the University of Hamburg. Part of the In Figure 1 we present the ProLiV System archi- lessons content was adapted following the user’s tecture, consisting of: feedback. • a ﬁle repository (lessons, dictionary, tests, and exercises), 2.1 Animations in ProLiV Animations are not integrated in all modules of • a tool repository, the ProLiV system, but only in the LSA, computa- • an aggregating module combining elements tional lexicons and question modeling modules. from ﬁle and tool repository (Main Unit), In order to decide how to organize the informa- tion in an animation, we evaluated the animations • the graphical user interface (G.U.I.) for the matrix multiplication in the LSA module For each topic a stand-alone module is con- by asking 11 high-school pupils (between 16 and nected with the G.U.I module via the Main Unit. 19 years old) to choose between the several repre- Modules related to new topics can be inserted any sentations. time with no particular changes of the system. We showed the pupils three animations that de- The ProLiV architecture follows the guideline scribe the multiplication of matrices, a static pic- considerations found in (Galitz, 1997). ture and the text representation of the deﬁnition. The animations differ in the way the process is presented (abstract vs. concrete) and in user in- teraction authorization. The pupils were asked to evaluate all the rep- resentations. The question they had to answer was: ”Which of the following representations helps more, when learning about matrix multipli- cation?”. The scale given was from 1 = very help- ful to 5 = not helpful at all. Analyzing the results, we could not conclude that one representation is a ”real winner’. The best representation was considered the most ﬂex- ible animation, that allows the student go back- wards and forwards whenever the user needs it, Figure 1: The ProLiV Architecture 14
The ﬂexibility of the system is also given by the 3.1 Integrated external software packages fact that the G.U.I.1 is generated according to an The learning process is also sustained by in- XML2 description, developed within the project teractive elements, such as the possibility of (see DTD Description). changing parameters for the LSA algorithm The XML description contains the information and visualizing the results, or as the inte- in the lessons (deﬁnitions, theory, examples, etc.) grated programs for the computational lexicons and the G.U.I. speciﬁcations (colors, fonts, links, tool: ManageLex (http://nats-www. arrangement in the interface, etc.). Having an informatik.uni-hamburg.de/view/ XML ﬁle as input, the system generates automat- Main/ManageLex) and G.E.R.L. (http:// ically the G.U.I. presented to the student. The in- nats-www.informatik.uni-hamburg. formation shown to the user can be extended or de/view/Main/GerLexicon). This way modiﬁed with almost no implementation effort. the students have the possibility, not only to read New lessons or modules can be integrated, by ex- the theory, but also to see the impact of their tending or adding XML ﬁles. Due to the same fact, modiﬁcations in an algorithm that is described in also the content adaptation of the system to other the lessons. languages3 is very easy. Due to its architecture, other such external pro- grams can be easily integrated within ProLiV. The DTD Description: 4 LSA Module in ProLiV The LSA module makes an introduction to the topic. It gives an overview of the LSA algo- rithm, principles, application areas, and of the main mathematical notions used in the algorithm. Initially thought for being used mostly by students from linguistics (or linguists) - due to the mathe- matical algorithms -, the tool can be exploited by ........................ anybody who wants to have an introductory course on LSA. The content is organized in four Units: 1. LSA: General Knowledge - It gives the LSA deﬁnition, a short overview of the history, its semantics, and how LSA can be used in the study of cognitive processes. 2. Mathematical Fundamentals - It describes the The G.U.I. follows the same design rules in all LSA algorithm modules and the layout and format decisions are 3. LSA Applications - It presents the applica- consistent. A color and a font style are associated tion areas for the LSA, LSA limitations and to only one kind of information (e.g. color red as- critics. Also a comparison with other similar sociated to deﬁnitions, etc.). algorithms is made. 1 The G.U.I. is automatically generated not only for the 4. Compendium of Mathematics - It gives the lessons, but also for the term dictionary associated to each user the mathematical background: deﬁni- module. tions, theorems, etc. 2 XML = Extensible Markup Language. More details to be found on http://en.wikipedia.org/wiki/XML 3 For the moment ProLiV contains lessons in German and The course has also an introduction, a motivation, English conclusion and references. 15
The LSA module is offering not only a textual the ProLiV homepage: http://nats-www. representation of the information, but also sev- informatik.uni-hamburg.de/view/ eral visualization methods (as images and anima- PROLIV/WebHome. tions4 ). Beside the lessons, there are implemented a term dictionary and an environment for testing Acknowledgments LSA parameters. We would like to thank all people that helped in the development of our software: Project Coor- 4.1 The LSA Test Environment dinator Prof. Dr. Walther v. Hahn (Computer Probably the most interesting part of the LSA Science Department, Natural Language Systems module is the test environment. After learning Group), Prof. Dr. Angelika Redder (Depart- about LSA, in this environment the user has the ment of Language, Linguistics and Media Stud- possibility to actually see how LSA is working, ies, Institute for German Studies I), Dr. Shinichi and what results can be obtained when compar- Kameyama (Department of Language, Linguistics ing the meaning of two words. The user can set and Media Studies, Institute for German Stud- several parameters of the algorithm - e.g. the ies I), Christina von Bremen (Computer Science analysis mode (simple/frequency based vs. ad- Department, Natural Language Systems Group), vanced/entropy based), the minimum word occur- Olga Szczepanska (Computer Science Depart- rences, the analysis dimension, the similarity mea- ment, Natural Language Systems Group), Irina sure (Cosine, Euclidean, Pearson, Dot-Product), Aleksenko (Computer Science Department, Nat- etc. - and decide which words are not considered ural Language Systems Group), Svetla Boytcheva in the analysis. The analyzed text, the initial co- (Academy of Sciences Soﬁa). occurrence matrix and the one obtained after ap- plying the Singular Value Decomposition (SVD) algorithm are shown in the G.U.I. The similarity References measure, when comparing two words, is calcu- Wilbert O. Galitz. 1997 The Essential Guide to User lated in both unreduced and reduced cases. Interface Design: an Introduction to GUI Design principles and Techniques, Wiley Computer Pub- lishing, New York. 5 Conclusions Robert J. Gaizauskas, Peter J. Rodgers, and Kevin The paper presents a course-ware software, Pro- Humphreys. 2001 Visual Tools for Natural Lan- LiV. It is a collection of (interactive) multimedia guage Processing, Journal of Visual Languages and tools used mainly for the consolidation of ﬁrst- Computing, Vol. 12, Number 4, p. 375-411, Aca- demic Press years courses in computational linguistics and lit- erary computing. Its goal is to help the humanist Monica Gavrila, Cristina Vertan. 2006 Visualization scientists to make use of complex formal methods, of Complex Linguistic Theories, in the Proceed- ings of the ICDML 2006 Conference, p. 158-163, and the computer specialists to understand human- Bangkok, Thailand, March 13-14 ist facts and interpretations. The main feature of the system, in the context Monica Gavrila, Cristina Vertan, and Walther von Hahn. To be published during 2009 ProLiV - Learn- of the conference, is not the content of the lessons, ing Terminology with animated Models for Visualiz- but the system’s extendible and adaptable architec- ing Complex Linguistics Theories, in the Proceed- ture. Another important aspect is the way in which ings of the LSP 2007 Conference, Hamburg, Ger- the information is presented to the student. many, August, The system runs on any platform supporting Julie Bauer Morrison, Barbara Twersky, and Mireille Java 1.5 or newer. It was developed on Linux and Betrancourt. 2000 Animation: Does It Facilitate tested on Windows and Mac OS X. Learning?, in the Proc. of the Workshop on Smart Graphics, AAAI Press, Menlo Park, CA. Being Java-based and having as input Unicode ﬁles (XML encoded information), the system can Kay L .Orr, Katharine C. Golas, and Katy Yao. 1994 be embedded in the future in a Web environment. Storyboard Development for Interactive Multimedia Training, Journal of Interactive Instruction Devel- More about ProLiV can be found in (Gavrila opment, Volume 6, Number 3, p. 18-31 et al, 2006) or in (Gavrila et al, TBA) and on Pete Thibodeau. 1997 Design Standards for Visual 4 The animations integrated are for the LSA algorithm Elements and Interactivity for Courseware, T.H.E. tested on an example and for matrix multiplication Journal, Volume 24, Number 7, p. 84-86 16