Arabic, International & Multilingual Software Desktop Publishing Machine Translation Document Management OCR ASR TTS MultimediaA

TranSphere™ Hybrid Machine Translation (MT)

TranSphere® is AppTek’s state-of-the-art Hybrid Machine Translation (MT) solution. TranSphere Hybrid Machine Translation is currently available for several languages. AppTek is actively migrating all supported languages to its Hybrid Machine Translation platform. TranSphere can be integrated with AppTek’s automatic speech recognition product, Plain Speech, and text-to-speech for a speech machine translation system loadable on PCs and wearable devices.

Machine Translation is the use of computer software to translate text from one natural (human) language into another. Machine Translation must take into account the grammatical structure of each language, and use contextual rules to select among multiple meanings, in order to transfer sentences from the source language (text to be translated) into the target language (translated text). AppTek developed TranSphere after several years of extensive linguistic research in the US, Asia and the Middle East. Its technology reaches beyond other, more simplistic approaches--some of which verge on mere word-by-word substitution—towards real-life, high-volume industrial and commercial applications.

TranSphere's® state-of-the-art computational-linguistic technology—for which APIs and SDKs are available—supports translation between English and a variety of other languages (bidirectional): Arabic, Farsi, Korean, Japanese, Chinese, Turkish, Persian\Dari, Urdu, Pashto-English, Bahasa Indonesian and Tagalog. The AppTek line of Machine Translation coverage also includes European languages: French, German, Italian, Portuguese, Polish, Russian, Spanish, Ukrainian, Hebrew and Dutch.

Components

There are two components for the Machine Translation system: an engine which processes the translation, and an environment which allows a user to submit text for translation, receive the results, and manipulate the text before and after the translation. This environment is compatible with a stand-alone Machine Translation workstation, Client/Server architecture and Web-based solutions.

TranSphere also supports MS Office, MS IE and includes an API for third-party integration. The Client/Server solution provides greater flexibility than a stand-alone workstation. The client environment can be ported to multiple platforms without impacting the server application. This provides the capability to tailor the user environment to a particular platform, using all of the commercially available products and tools that are inherent to that platform. Servers can be Windows NT, UNIX, SUN, SCO, LINUX, etc. Since a Machine Translation engine is both CPU and memory intensive, the selection of a hardware platform should focus on disk, memory, CPU performance; expandability; commercial availability and in-country support. The platform should also have the capability to support multiple processors to provide increased performance. Requirements for the client platform are much less stringent. Any standard desktop platform, which supports a windowed environment, is suitable.

Approach

TranSphere has been designed and implemented by utilizing natural language high-level linguistic programming tools in the Lexical Functional Grammar paradigm. The programming language has been used to create data tables containing linguistic symbol definitions, lexical entries, linguistic rules (for analysis, transfer, and generation), and other control tables, which are then compiled into the file system. The high-level programming language is compiled using the ANSI C programming language, which provides a robust and powerful tool for the computational linguist to focus efforts on Machine Translation development while providing maximum portability across various platforms and operating systems. The system provides integrated hardware and software solutions that will:

• Analyze semantic, morphological and syntactic structures and produce the output translation.
• Function as a fully automated Machine Translation system.
• Utilize a general-purpose lexicon.
• Access special domains micro-dictionaries pertaining to such domains as Computer, Aviation, Military, Telecommunications, Economics, Diplomacy, Nuclear, Drug, Chemical, Law Enforcement and Business.
• Provide users with options to define the stacking of those dictionaries.
• Support a workstation-based solution with a client-server architecture.
• Support a Vocabulary Search Utility to identify words not in the system's dictionaries.
• Support a Dictionary Maintenance Utility to add user-defined words.
• Produce Latin-character-dropped words for those unfound in dictionary.
• Interface with a Text Management System.
• Interface with an OCR and a spell checker.
• Interface with an ASR and TTS to produce Speech-to-Speech Machine Translation

These tools can be modules in an integrated translation and editing environment compatible with existing platforms, email and office applications. They are part of the overall environment in the English source as well as the telecommunication needs of a bilingual translation, editing and electronic publishing in a global multi-site organization. The issues pertaining to the timeliness, quality of translated content and the productivity of the translation and editing teams will be optimally enhanced with the use of Machine Translation tools like TranSphere.

The Hybrid Approach

Compared with written language, speech (especially when spontaneous) poses additional difficulties for the task of automatic Machine Translation. Typically, these difficulties are caused by errors of the recognition process, which is carried out before translation. As a result, the sentence to be translated is not necessarily well-formed from a syntactic point-of-view. Even without recognition errors, speech translation has to cope with a lack of conventional syntactic structures because the structures of spontaneous speech differ from those of written language.

A prime motivation for a hybrid Machine Translation system is to take advantage of the strengths of both rule-based and statistical approaches, while mitigating their weaknesses. Thus, for example, we want a rule that covers a rare word combination or construction to take precedence over statistics that were derived from sparse data (and thus not very reliable). Additionally, rules covering long-distance dependencies and embedded structures should be weighted favorably, since these constructions are more difficult to process in statistical Machine Translation.

Conversely, we would like a statistical approach to take precedence in situations where large numbers of relevant dependencies are available, novel input is encountered or high-frequency word combinations occur. An aspect that is extremely important in regards to the distillation engine is the weakness that statistical Machine Translation sometimes has in informativeness (the accurate translation of information) due to the influence of the target-language model. For example, single words that may make a disproportionately heavy contribution to informativeness, such as terms indicating negation or important content words, may be missing.

Statistical Machine Translation Module

Our statistical Machine Translation is a finite state transducer using alignment templates. Compared to traditional statistical MT systems, these methods have the advantage of being capable of learning translations of phrases, not just individual words, which permits the MT to encompass the functionality of example-based approaches and translation memories. The other advantage is that it allows for the combination of many knowledge sources, by framing them as feature functions that are combined using a Maximum Entropy framework.

Rule-Based Machine Translation Module

Our rule-based module employs a Lexical Functional Grammar (LFG) system. The LFG system contains a richly-annotated lexicon containing functional and semantic information. It also produces richly-annotated intermediate outputs that may interact with the statistical MT module:

• Source language c-structure (or "constituent structure") – a phrase-structure tree
• Source language f-structure (or "functional structure") – a directed acyclic graph containing attribute-value pairs specifying grammatical information such as grammatical relation (subject, object, etc.), case, argument structure (predicate, argument, adjunct), semantic disambiguation information (human, animate, concrete, etc.), gender, number, mood, tense, aspect, polarity, speech act, etc.
• Target language f-structure – an f-structure that has been modified and restructured to enable generation of target-language text

Features

• Capacity and Speed
  • Human translator can translate about 2,000 to 3,000 words a day. TranSphere translates thousands of words per minute.
• Legacy Data
  • TranSphere allows for the re-use and storage of previously translated documentation, in whole or in part utilizing a statistical module Translation Memory.
• End-user Products
  • TranSphere is available in several versions:
    • Standalone
    • Client/Server
    • Web Trans Server

Cost

• Human translation costs anywhere from US $0.20 to US $0.60 per word. TranSphere pays for itself within the first year of use.

TranSphere® is integrated with:

• Optical Character Recognition (OCR)
• Automatic Speech Recognition (ASR)
• Text-to-Speech (TTS)
• Speech-to -Speech MT (Both ASR & TTS)

TranSphere also has the following optional add-ons:

• Office + MS-web translator
• Third Party Integration API
• Term Substitution

Please, contact George N. Hallak for Multi-License pricing and customized solutions