View
467
Download
1
Embed Size (px)
DESCRIPTION
Citation preview
Dynamic Syllabi for Historical Language Instruction �
Digital Humanities, University of Leipzig
Dynamic Syllabi for Historical Language Instruction
Part 1: Globalization and Localization Its relevance to historical languages and resulting challenges.
Part 2: User Experience for eLearning
The resulting web interface and user experience for learners. Part 3: Games, Graphs, and Data
eLearning games: the data and system that drives it.
2
Dynamic Syllabi for Historical Language Instruction
Part 1: Globalization and Localization Supporting languages of the local learners
Part 2: User Experience for eLearning The resulting web interface and user experience for learners.
Part 3: Games, Graphs, and Data
eLearning games: the data and system that drives it.
3
Dynamic Syllabi for Historical Language Instruction
Part 1: Globalization and Localization Supporting languages of the local learners
Part 2: User Experience for eLearning Language independent functions
Part 3: Games, Graphs, and Data
eLearning games: the data and system that drives it.
4
Dynamic Syllabi for Historical Language Instruction
Part 1: Globalization and Localization Supporting languages of the local learners
Part 2: User Experience for eLearning Language independent functions
Part 3: Games, Graphs, and Data
How do you build the backend?
5
Latin and Greek
Latin and Greek are taught in across Europe Primary and secondary school instruction in the national language à at least 24 national languages within Europe alone….
6
Globalization
The process of making all the necessary technical, financial, managerial, personnel, marketing and other enterprise decisions necessary to facilitate international business.
Being global = Providing materials for each language that are suitable for learning historical languages based on their native language.
7
“Omni-local” instead
Respecting and enhancing local cultures
and variation.
8
Perseus Digital Library
Scaife Digital Library
Historische Sprachen eLearning
Projekt
Why be Global?
9
Perseus Digital Library
Scaife Digital Library
Historische Sprachen eLearning
Projekt
Why be Global?
10
Perseus Digital Library
Scaife Digital Library
Historische Sprachen eLearning
Projekt
Internationalization and Localization
11
Internationalization is the process of enabling a product at a technical level for localization. Localization is the process of modifying products or services to account for differences in distinct markets. Source: LISA (The Localization Industry Standards Association)
Localization: Linguistic Issues
12
Adaptation of the content for Croatian and Persian speakers, Comparison:
• Explaining, what is a Dative case for Persian speakers !== Croatian speakers don’t need this, because they have 7 cases in their language (Bulgarians have none!), BUT Croatians don’t have a definite article.
• No need for explaining the function of participles for Persian speakers !== Croatians need to know, what a participle is.
Almost any product or service that will be sold to individuals who do not speak the language in which it was created will require linguistic adaptation.
Localization Challenges
13
Physical Issues Beyond translation, localization often involves physical modification to products or services in order to be acceptable in the local market.
Business and Cultural Issues
Local business and cultural issues can affect all aspects of product design and localization: e.g. numbers, names, colors and graphics.
C = L
Technical Issues Supporting local languages may require special attention and planning at the engineering stage: e.g. right to left direction, date formats, separators in the numbers.
Europe: 03.12.2013
US: 12/03/2013
Islamic Countries: 29.01.1435
Iran: 1392/09/12
Chinese Calendar: Jia-Zi(Rat) (11th month), 1, 4711
Technical Issues: Date Formats
14
Interlingua: L1-independent display of a sentence
15
But you can’t avoid the L1 of the learner!
16
• You cannot avoid translating L2 to L1 during learning the language
• Translation helps dynamic learning
• Using translation doesn’t mean, going back to the grammar-translation method
• It is not a learning method itself, but it could be combined with other methods.
Translation strategies in the learning process
Literal English
Literary English
17
Literal Persian
Translation: Clarification
18
• We are not talking about literary translations (i.e., free translations that capture the spirit of the original but do not follow the original closely).
• The purpose of translation is to learn and to demonstrate what you have learned – more literal, more applied
• Translation is the skill to be used to develop language understanding
• We also DO need a lot of new translations in many languages
• We are doing collaborative translation by named individuals, not an anonymous crowd.
• We need a FIRST direct translation of Plato’s Republic into Persian.
Question
19
Knowing other languages is not always a good point:
The help systems are so good that you can translate without learning (e.g., you have aligned Greek/English, morpho-syntactic annotations,
dictionaries, commentaries and then you translate into Persian!)
How do you internalize knowledge of the language?
eLearning User Experience (UX)
20
Localization and a graph-based backend are both
important components that ultimately make our goal
eLearning user experience possible
What exactly is UX?
21
User Experience includes…
Usability
System Performance
Accessibility Interaction Design
Utility Graphic Design
21
UX for eLearning
22
eLearning presents some interesting user experience challenges such as:
• Improve understanding and retention of learning materials
• Teach users novel interactions required for novel learning tasks
like treebanking and alignment
• Accommodate the wide variety of learning goals for different
types of users based on their interests
UX for eLearning
23
eLearning also presents some interesting user experience opportunities such as:
• Personalize a user’s learning experience; go beyond customization
• Provide detailed and immediate feedback for users based on their
responses to exercises
• Visualize user progress in a way that shows how what they have
learned maps directly to their target corpus
Acquainting the User with our System
24
• Based on a traditional Greek textbook (John William White’s First Greek Book) our learning materials are divided into lessons.
• Immediately gives the user a sense of place, and progress as it relates to Ancient Greek grammar. The interface clearly communicates, “Start Here.”
Providing Goals and Feedback
25
• Show a user what they’ll accomplish in the lesson.
• As the user progresses through the lessons, they see that the things they learn are directly related to their target corpus.
Visualize Progress within the Target Corpus
26
• Since the system itself is optimized for a target text, it quickly becomes clear how a relatively small number of vocabulary words and grammar rules helps a user make huge strides in learning in a short time.
New Interactions for New Kinds of Learning
27
• We start slowly to introduce the concept of treebanking.
• Provide feedback while the user is building the tree, until they are comfortable with the new interaction.
• Give the user specific corrections once they’ve submitted an answer.
Enhancing the UX going forward
28
• Use recorded metrics to discover the ways people learn and retain information most effectively.
• Personalize the interface and experience more acutely. • Use richly annotated text to provide numerous examples of grammatical
constructions and vocabulary words in context.
• Provide further texts, from which users can learn.
Goal
Interactive and dynamic learning + more and better feedback for students
Games cover every stage in the workflow of a digital edition
Linguistic Annotation Aligned Translation Transcription + Structural Markup
Identifying Named Entities
29
Motivation
Linguistic Annotation �
Aligned Translation
Transcription�+ Structural
Markup
Identifying Named Entities
• Identify the morphology of a given word and context
• Identify the syntactic function of a word (treebanking)
• Fill in missing word (forms)
• Align new translation
• Suggest correction for existing translations
• Practise typing by Captchas
• Identify OCR errors
• Who/where/what is it?
• Uncover ethnicities, locations, events in ancient texts
30
Data – Intersection�
31
Francesco Mambrini
Bruce Robertson, Federico Boschetti
Leif Isaksen, Gabriel Bodard
Data Preprocessing
32
Syntax
Morphology
Alignment
Preprocessing/ Format Normalization STORAGE
RDB – Why not?
33
Modelling this ER model as RDB schema means: • 1 table per entitiy and • 1 table per relationship → at least 30 (gave up after 7 tables) • Adding new model components means: either rebuild the db or put high effort into persistence and
integrity
Data Preprocessing
34
STORAGE
GRAPH
Representation
Why Graphs?
35
What:
• Entities and relationships
• Nodes and the way they relate (to the world) to each other as edges
How:
• Scalable
• Additive
Entity
Entity
Entity
Relationship
Relationship
Graph Performance
Performance stays stable when dealing with highly connected data
RDBs then require join-intensive queries where performance slows down with growing dataset
Not so with graphs, because queries are localized to a portion of graph traversed to satisfy that query
36
Graph Performance
37
- traversals (shortest paths, exists a path, etc.) are more performant than in RDBs (huge joins) - existence of well-performing algorithms (e.g. Dijkstra) on graphs
Depth RDBS exec. time Neo4j exec. time Records returned
1 0.016 0.01 ~2500
2 30.267 0.168 ~110,000
3 1543.505 1.359 ~600,000
4 Unfinished 2.132 ~800,000
from: Ian Robinson, Jim Webber, Emil Eifrem. Graph Databases. O'Reilly Media. 2013. p. 20
Benchmarks
38
Depth Neo4j exec. time Records Returned
2 ~0.5 ~200
2 ~0.6 ~500
2 ~0.8 ~ 5,000
Client: Virtual Client Ubuntu 12.0.4 on (1 core, 2 GB RAM from the host) Host: Windows 7 Professional Intel Core i5-2400 CPU 2 cores á 3,10 Ghz, 4GB RAM
Data Additivity
39
Add new kinds of relationships, nodes & subgraphs to existing structure without affecting application functionality.
Data Additivity
40
Add new kinds of relationships, nodes & subgraphs to existing
structure without affecting application
functionality.
40
Data Queries
41
Return every word with the “POS” property “noun”
START doc=node(*)
MATCH (doc)-[:CONTAINS_SENT]->(sent)-[:CONTAINS]->(w)
WHERE HAS (w.pos) AND w.pos=“n”
RETURN DISTINCT w.cts sentence
word document
contains
Data Queries
42
Return every word of a sentence that contains at least one word with the POS property “verb”
START s=node(*)
MATCH (s)-[:CONTAINS]->(w)
WHERE HAS (w.pos) AND w.pos=“v”
WITH s
MATCH (s)<-[:BELONGS_TO]-(w2)
RETURN s, w2 ORDER BY w2.cts ASC
Data Queries
43
Return every word of a sentence that contains at least one word with the POS property “verb” learned by the user “John” during the first week of the semester.
START s=node(*)
MATCH(s)-[:CONTAINS]->(w)<-[:CONTAINS]-(submission)<-[:SUBMITTED]-(u)
WHERE HAS (w.pos) AND w.pos=“v” AND u.name=“John” AND submission.time <
24.1.2014
WITH s
MATCH (s)<-[:BELONGS_TO]-(w2)
RETURN s, w2 ORDER BY w2.cts ASC
Thank you�
Questions?
44