Interlingual Technology Overview

Notes from a Presentation to Esperantic Studies Foundation

Christie Evans and Jonathan Robert Pool
2003/03/23
Revised 2004/07/22

World trends in language-related technology

What is interlingual technology?

Fields of interlingual technology

World trends in language-related technology

Growth of information-processing power:

Growth of Internet use:

Growth of language diversity on the Internet:

Recent changes:

Current distribution:

What is interlingual technology?

Candidate Definition

Interlingual technology is technology which by design or effect promotes or facilitates communication that, in the absence of such technology, would be prevented or impaired by language diversity.

Main fields:

• Translation technology
• Localization
• Language learning
• Interlingual design

Fields of interlingual technology

Translation technology

Problem

Prototypical purpose: Render A's production of content into language known by B so B can consume it without A or B learning any language not already known.

Obstacles to (automatic) translation:

• Grammatical complexity
• Vagueness
• Polysemy
• Ambiguity
• Multi-purpose content
• Multiple incompatible criteria of content quality
• Inter-language incompatibility
• Ineffability
• Competing criteria of translation performance, e.g. remove ambiguity versus reproduce ambiguity
• Dissensus on translator role: whisperer, mediator, editor
• Variation in translation aspiration: from topic to gist to content

Solution strategies

Notions of translation automation:

• Fully automatic transation
• Human-aided machine translation
• Machine-aided human translation
• Translation decision support
• Teletranslation
• Translation access automation

Overlapping automation methods include:

• Rule-based
• Corpus-based
• Example-based
• Statistics-based
• Knowledge-based
• Interlingua-based
• Transfer-based

These methods differ in whether translator attempts to "understand" the original.

Examples of problems

Polysemy (see Smith, TMRL, 1989, 86-87): "Se vi estas esperantisto, vi devas paroli Esperanton." Translatable into English as:

• alethic: "If you are an Esperantist, by definition you speak Esperanto"
• epistemic: "If you are an Esperantist, I'm confident that you speak Esperanto"
• deontic: "If you are an Esperantist, you have no excuse not to speak Esperanto"
• polysemic: "If you are an Esperantist, you must speak Esperanto"

Ambiguity (see Hutchins, MTTASPUP, 2002):

• "The hallucinating magicians saw the woman in two pieces in a box."
• "The authorities encouraged smog and pollution control."
• "light: léger, clair, facile, allumer, lumière, lampe, feu"

Ineffability?

• Turkish "ne zaman düşeceğini sorduğum ağaç" = English "*the tree that I asked when was going to fall"
• English "Do you remember the game that I told you Ronald thought his brother would never learn how to play?" = German "*Erinnern Sie sich am Spiel, daß ich Ihnen gesagt habe, daß Ronald gedacht hat, daß sein Bruder nie zu spielen lernen sollte?"

State of the art

Consensus: Nothing exists that approximates fully automatic high-quality translation of content of any kind about anything in any language into any other language.

Internet-based translation services: FreeTranslator. Examples:

Progress forecasts

How to operationalize progress. Reduction in cost of high-quality translation?

Consensus: There is no basis for a forecast when, if ever, fully automatic high-quality translation of content of any kind about anything in any language into any other language will exist.

Localization

Problem

Prototypical purpose: Make an information system usable by anybody, so that the user experience is equivalent, regardless of the languages and conventions known and preferred by the user.

Why localization is hard:

• It requires translation.
• The required translation can be incremental.
• The content to be translated can be user-generated.
• The required translation can be instantaneous.
• It requires adaptation for multiple convention types.
• Translations and convention adaptations constrain design decisions.
• The automated production of content in a language requires prior analysis and engineering.
• The initial design of information systems is localizationally myopic.
• Users may select and configure the delivery equipment, which may be unknown to producers and is partly unmodifiable.

Problem example: Can letter case be manipulated language-independently?

• Languages sharing scripts don't share case mappings (e.g., "i" & "I" in English, "i" & "İ" in Turkish).
• Languages with case don't share case rules.
• Diacritical marks are required on upper-case letters in some languages and not in others.
• Some languages have only 1 case.

Solution strategies

Localization notions:

• Globalization
• Localizability
• Customizability
• Internationalization
• Multilingual standardization
• Layering

Styles of localization:

• Uniformation
• Differentiated syndromic control
• Formal surrender

State of the art

Recommendations for localized software development: "Globalization Step-by-Step".

The Translation Project: "aims to get the maintainers of free software packages, benevolent translators, and users all together, so that computers and systems (like Linux or GNU) gradually become able to speak many languages." Results tabulated in Translation Matrix.

Example of an internationalized Web site

Unicode character pallette.

Unicode in 9 browsers.

Indicator of status quo: 10 firms pay for listings on a major search engine under the search string "multilingual Web site".

Progress forecasts

Is localization a norm? Is it increasingly a necessity?

The capability to display web sites in multiple languages is growing in importance every day. Two years ago almost the entire Internet spoke English as a first language. Today less than half the Internet speaks English as a first language. In another year this figure will be even less. If you were telephoning someone in another country, you would know that to communicate accurately you would have to speak their language. This same simple fact is true for all web sites. (http://www.c-prompt-dev.com/desc_webpage_services.asp#Internationalization)

Surveys show that Internet users are twice as likely to stay at a given site and three times as likely to make inquiries over the Internet if the content of the site is communicated in their own language. (http://www.globallisting.com/globallisting/showcase.html)

Motivation: Is it really profitable to market Jeep Cherokees in Cherokee?

Language learning

Problem

Prototypical purpose: Give A competence in a language so A may produce and consume content in it, thereby avoiding translation.

Solution strategies

Notions of language-learning technology:

• Language analysis
• Learner analysis
• Performance requirements analysis
• Tutorial automation
• Tutorial individualization
• Virtual immersion
• Remote immersion
• Application of general instruction technologies to language instruction

State of the art

Consensus: The application of information technology and artificial intelligence to language learning has not yet substantially improved the competence achievable per unit of investment in group instruction, but has substantially decreased the cost barriers to isolated, on-demand, and specialized language learning.

Web-based instructional support (Esperanto)

Web-based language learning (Esperanto)

Progress forecasts

I need, therefore I will?

Sleep learning? "Walk in, promenu out?"

Interlingual design

Problem

Prototypical purpose: Design or modify a language so it promotes or facilitates communication that is prevented or impaired by language diversity.

Language types

Regularized novel synthetic human language.

Disambiguated version of existing human language.

Automated controlled content-elicitation system.

Effect types

Translation interlinguas

Translation input media

Localization canonical representations

Learning appetizers

Learning objects

Progress forecasts

Controlled input progressing because of recognized value.

Use of designed human language for learning promotion may depend on ESF.