Machine translation and telecommunications system using user ID data to select dictionaries

Description Submit all comments and votes

SPECIFICATION

1. Technical Field

This invention relates to a system for automatic (machine) translation of text and, more particularly, to a telecommunications-based system for automatically translating text in a source language to a target language using user ID data to select machine dictionaries used by the system.

2. Background Art

After several decades of development, the field of automatic (machine) translation of text from a source language to a target language with a minimum of human intervention has developed to a rudimentary level where machine translation systems with limited vocabularies or limited language environments can produce a basic level of acceptably translated text. Some current systems can produce translations for unconstrained input in a selected language pair, i.e., from a chosen source language to a chosen target language, that is perhaps 50% acceptable to a native writer in the target language (using an arbitrary scale measure). When the translation system is constrained to a particular vocabulary or syntax style of a limited area of application (referred to as a "sublanguage"), the results that can now be achieved may approach a level 90% acceptable to a native writer. The wide difference in results is attributable to the difficulty of producing accurate translation when the system must encompass a wide variability in vocabulary use, syntax, and expression, as compared to the limited vocabularies and translation equivalents of a chosen sublanguage.

One example of a machine translation system limited to a specific sublanguage application is the TAUM-METEO system developed by the University of Montreal for translating weather reports issued by the Canadian Environment Department from English into French. TAUM-METEO uses the transfer method of translation, which consists basically of the three steps of: (1) analyzing the sequence and morphological forms of input words of the source language and determining their phrase and sentence structure, (2) transferring (directly translating) the input text into sentences of equivalent words of the target language using dictionary look-up and a developed set of transfer rules for word and/or phrase selections; then (3) synthesizing an acceptable output text in the target language using developed rules for target language syntax and grammar. TAUM-METEO was designed to operate for English-to-French translation in the narrow sublanguage of meteorology (1,500 dictionary entries, with several hundred place names; text having no tensed verbs). It can obtain high levels of translation accuracy of 80% to 90% by avoiding the need for any significant level of morphological analysis of input words, by analyzing input texts for domain-specific word markers which narrow the range of choices for output word selection and syntax structure, and by using ad hoc transfer rules for output word and phrase selections.

Another example of a sublanguage translation system is the METAL system developed by the Linguistics Research Center at the University of Texas at Austin for large-volume translations from German into English of texts in the field of telecommunications. The METAL system also uses the transfer method, but adds a fourth step called "integration" between the analysis and transfer steps. The integration step attempts to reduce the variability of output word selection and syntax by performing tests on the constituent words of the input text strings and constraining their application based upon developed grammar and phrase structure rules. Transfer dictionaries typically consist of roughly 10,000 word pairs. In terms of translation quality, the METAL system is reported to have achieved between 45% and 85% correct translations.

A strategy competing with the transfer approach is the "interlingua" approach which attempts to decompile input texts of a source language into an intermediate language which represents their "meaning" or semantic content, and then convert the semantic structures into equivalent output sentences of a target language by using a knowledge base of contextual, lexical, and syntactic rules. Historically, transfer systems lacking a comprehensive knowledge base and limited to translation of sentences in isolation have had the central problem of obtaining accurate word and phrase selections in the face of ambiguities presented by homonyms, polysemic phrases, and anaphoric references. The interlingua approach is favored because its representation of text meaning within a context larger than single sentences can, in theory, greatly reduce ambiguity in the analysis of input texts. Also, once the input text has been decompiled into a semantic structure, it can theoretically be translated into multiple target languages using the linguistic and semantic rules developed for each target language. In practice, however, the interlingua approach has proven difficult to implement because it requires the development of a universal symbolic language for representing "meaning" and comprehensive knowledge bases for making the conversions from source to intermediate and then to target languages. Examples of interlingua systems include the Distributed Translation Language (DLT) undertaken in Utrecht, the Netherlands, and the Knowledge-Based Machine Translation (KBMT) system of the Center for Machine Translation at Carnegie-Mellon University.

Other machine translation systems have been developed or are under development using modifications or hybrids of the transfer and interlingua approaches. For example, some systems use human pre-editing and/or post-editing to reduce text ambiguity and improve the correctness of word and phrase selections. Other systems attempt to combine a basic transfer approach with knowledge bases and artificial intelligence techniques for machine editing and enhancement. Another approach is to combine decompilation to a syntactically-based intermediate structure with transfer to equivalent output phrases and sentences. For a further discussion of current developments in machine translation, reference is made to Machine Translation, Theoretical and Methodological Issues, edited by Sergei Nirenberg, published by Cambridge University Press, 1987, and "Proceedings of The Third International Conference on Theoretical and Methodological Issues in Machine Translation of Natural Language", published by Linguistics Research Center, University of Texas at Austin, June 1990.

It is expected that machine translation (MT) systems will develop in time to provide higher levels of translation accuracy and utility. However, current MT techniques using a basic transfer approach can produce acceptable translation accuracy in a selected sublanguage, yet they are not in widespread use. One reason for the limited use of MT systems is that most current systems are designed for a single, specific application, environment and language pair context. The requirements of that context motivate the design and development of the grammar, dictionary structure, and parsing algorithms. Thus, the utility of the system becomes confined to that particular context. This approach greatly limits the range of applications and the audience of users which can be productively served by such application- and language-specific MT systems.

SUMMARY OF INVENTION

It is therefore a principal object of the present invention to provide a system for performing machine translation for different source languages, target languages, and sublanguages, and automatically sending the translated text via telecommunications links to one or more recipients in different languages and/or in different locations. The system should be capable of providing acceptable levels of translation accuracy and be readily upgradable to higher levels of accuracy and utility. It is a further object that such a system be capable of operation with a minimum of human intervention, yet have interactive utilities for obtaining and adding new word entries to its dictionary database. It is also desired that such a system be capable of building and organizing a large-scale dictionary database containing core language dictionaries, plural sublanguage dictionaries, and individual user dictionaries in a manner which cumulates and evolves over time.

In accordance with a principal aspect of the present invention, a machine translation and telecommunications system comprises:

(a) a machine translation module for performing machine translation from input text of a source language to output text of a target language;

(b) a receiving interface for receiving input via a first telecommunications link, said input including an input text to be translated accompanied by a control portion having at least a first predefined field therein for designating an address of a recipient to which translated output text is to be sent;

(c) a recognition module coupled to said receiving interface for electronically scanning the control portion and recognizing the address of the recipient designated in the first predefined field of the control portion; and

(d) an output module including a sending interface for sending translated output text generated by said machine translation module to the address of the recipient recognized by said recognition module via a second telecommunications link.

In a more specific aspect of the invention relating to sublanguage selection, a machine translation system comprises:

(a) a receiving interface for receiving an input text and a sublanguage control input indicative of a selected sublanguage applicable to the input text from among a plurality of possible sublanguages;

(b) a machine translation module capable of performing machine translation of an input text in a source language to an output text in a target language using a dictionary database containing entries for words of the target language corresponding to words of the source language;

(c) a dictionary database including a core language dictionary containing entries for generic words of the source and target languages, and a plurality of sublanguage dictionaries each containing entries for specialized words of a sublanguage;

(d) a dictionary control module responsive to the sublanguage control input for selecting a sublanguage dictionary of the dictionary database applicable to the input text, and for causing the machine translation module to use the selected sublanguage dictionary in performing translation of the input text; and

(e) an output module for outputting translated text in the target language generated by the machine translation module.

In the present invention, the sublanguage control input causes a selected sublanguage dictionary deemed applicable to the input text to be used in order to perform more accurate translation of the input text. The dictionary database includes core and sublanguage dictionaries for different source/target languages and sublanguages. The machine translation system with this multiple core languages and sublanguages capability is employed in a telecommunications system which automatically translates and transmits text from a sender to one or more recipients in other languages. A cover page or header accompanying the input text is used to designate the selected source/target languages, the applicable sublanguages, and the address(es)--electronic, fax, or mail--of the recipient(s).

The output module may have a page formatting program for composing the translated output text into a desired page format appropriate to a particular recipient or target language. It may also have a footnoting function for providing footnotes of ambiguous phrases of the input text in their original source language and/or with alternate translations in the target language. The output module includes a sending interface coupled dictionary database, a machine translation system comprises:

(a) a machine translation module for performing machine translation of input text in a source language to output text in a target language using a dictionary database containing entries for words of the target language corresponding to words of the source language;

(b) a dictionary database including a core language dictionary containing entries for generic words of the source/target languages, a plurality of sublanguage dictionaries each containing entries for specialized words of a sublanguage used by a group of users, and a plurality of user dictionaries each containing entries for individualized words of a user; and

(c) a dictionary control module responsive to control inputs to the machine translation system for causing the machine translation module to use the core language dictionary, any applicable sublanguage dictionary, and any applicable user dictionary for performing translation of an input text attributed to a user of the system.

In the invention, a large-scale dictionary database is maintained which has dictionaries containing word entries specified linguistically at different hierarchical levels of usage. At the lowest (user) level, a particular user can enter temporary or "scratch" word entries into a respective user dictionary. The machine translation system uses the particular user's dictionary to perform machine translation of text which may contain idiosyncratic or new words or phrases particularly used by that user. The dictionary control module includes dictionary maintenance utilities which allow such scratch entries to be entered by users into their user dictionaries, and which assist a dictionary maintenance operator (DMO) to review the scratch entries so that they can be confirmed as valid dictionary entries for machine translation. The dictionary maintenance utilities include automated programmed procedures for assessing whether word entries appearing in lower-level dictionaries should be moved into higher-level dictionaries.

Other objects, features, and advantages of the present invention will become apparent from the following detailed description of the preferred embodiments of the invention, as considered with reference to the following drawings:

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 is a schematic diagram of a machine translation and telecommunications system in accordance with the invention.

FIG. 1A is a schematic diagram of a computer server which includes a receiving interface, recognition module, and dictionary control module, and is coupled to a machine translation module and an output module.

FIG. 1B is a schematic diagram of a machine translation module which includes a translation processing module and a dictionary database, and its linkage to the computer server and the output module.

FIG. 1C is a schematic diagram of the output module, including a page formatting module and a sending interface.

FIG. 2 is an illustration of a cover page for designating core language pair, sublanguage(s), and recipient information, and accompanying text pages.

FIG. 3 is an illustration of input ideographic text and output English text as performed by the machine translation system using page formatting functions.

FIG. 4 is a schematic diagram of the dictionary control module, including dictionary selection and maintenance submodules, the latter containing an (interactive) user maintenance module and a dictionary maintenance module.

FIG. 5 is a schematic representation of an interactive input editor for interactions with users of the system.

FIG. 6 is a schematic diagram illustrating dictionary maintenance utilities for collapsing and promotion of entries from subordinate to superordinate dictionaries.

FIG. 7A illustrates, as a function of the dictionary maintenance utilities, the creation of scratch word entry from an identical word entry.

FIG. 7B illustrates the use of utilities with an interactive input editor to scan various levels of the dictionary hierarchy for word entries on which to base scratch word entries.

FIG. 7C illustrates a typical content of an identical word entry from which a scratch word entry is created.

FIG. 7D illustrates the creation of a "copy-cat" word entry from a synonymous word entry.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

Referring to FIG. 1, a preferred form of the machine translation and telecommunications system in accordance with the present invention comprises a computer server 10, a machine translation module 20, and an output module 30. (These and further-described components of the system will be denoted with capital letters for clarity of reference.) The Computer Server 10 receives electronic text input accompanied by a cover page or header from any of a plurality of input sources, designated generally as a telecommunications link A. The Computer Server 10 has a function for recognizing control data in the cover page or header designating core language and sublanguage selections applicable to the input text to be translated. It also recognizes output addresses and page formatting data to be used by the Output Module 30 for transmitting the translated text to the designated recipient(s) via any of a plurality of output devices, designated generally as a telecommunications link B. Due to the modularity of the system, the Machine Translation Module 20 may be updated by operator maintenance or upgraded or replaced without rendering the other functions of the system dysfunctional or obsolete.

The Machine Translation Module 20 is capable of performing machine translation from input text in a source language to output text in a target language. In the examples of a machine translation (MT) system described herein, reference is made to an MT system of the transfer type which relies upon the use of a machine-readable dictionary for lookup of source/target word entries. The principles of the present invention may also be applied to an MT system of the interlingua type. Transfer-type MT systems are widely accepted for near-term usage than interlingua systems, and they rely more heavily on linguistic knowledge incorporated into machine dictionaries designed for source/target language pairs. The operation of transfer-type MT systems is well understood by those skilled in the machine translation field, and is not described further herein.

Input Data Reception and Extraction

FIG. 1A shows the Computer Server 10 having a Receiving Interface 11 linked to the telecommunications link A, a Recognition Module 12, and a Dictionary Control Module 13. The Receiving Interface 11 may include an interactive mode program (to be described further herein) whereby a user can provide cover page or header designations, update or create User ID files pertinent to translation parameters associated with that user's communications, or create specialized user dictionary entries during interactive text entry sessions. The Recognition Module 12 includes a character recognition (often referred to as "OCR") program which recognizes and converts page image data into machine-readable text, and which recognizes cover page designations or user designations referencing cover page data stored in the User ID files. The Dictionary Control Module 13 includes a Dictionary Selection Module, which assesses the control data it receives from the Recognition Module 12 and designates the appropriate core language and sublanguage dictionary(ies) to be used by the Machine Translation Module 20. It also includes a Dictionary Maintenance Module, which allows a dictionary maintenance operator (DMO) to create and update dictionary entries in the Dictionary Database 22.

Using the control data from a cover page or header accompanying the input text, the Computer Server 10 allows the system to automatically recognize a sender's designations of the source language of the input text, the target language(s) of the output text, any particular sublanguage(s) used in a specialized domain, user group, or correspondence type, any preferred page format for the output text, and the address(es) of one or more recipients to whom the output is to be sent. Thus, the system can automatically access designated core and sublanguage dictionaries maintained in the Dictionary Database 22 for different source/target languages and sublanguages, and can format and transmit the translated text to recipient(s) in respective target language(s) via telecommunications link B, without the need for any substantial human intervention.

The Computer Server 10 interfaces with a plurality of receiving devices. For example, input data can be received as a facsimile transmission via a fax/modem board plugged into the I/O bus for the server system. Such fax/modem boards are widely available and their operation in a server system is well understood by those skilled in this field. Input may also be received from a conventional facsimile machine coupled to a telephone line which prints facsimile pages converted from signals transmitted on the telephone line. A conventional page scanner with a sheet feeder can be used to scan in facsimile or printed pages as page image data for input to the Computer Server. The page image data is then converted to machine-readable form by the OCR program. Input may also be received through a telecommunications program or network interface as electronic text or text files (such as ASCII text), in which case conversion by the OCR program is not required.

The OCR program may be resident as an application program in the Computer Server 10 along with the interface programs for handling the reception of input data. OCR programs are widely available, and their operation is well known in this field. For example, an OCR program for recognizing Japanese kana and ideographic characters is offered by Catena Corp., Tokyo, Japan. An example of an OCR program for alphanumeric characters is WordScan.TM. offered by Calera Recognition Systems, Santa Clara, Calif. The Computer Server 10 is preferably a high-speed, multi-tasking PC computer or workstation.

Referring to FIG. 2, the Computer Server 10 receives input data which is divided into two parts: a cover page or header 50 and input text 60. In the example shown, a cover page is used in conjunction with other pages of input text in a page-oriented system. In the case of transmission of an electronic text file or a text message, a preceeding header or identifier for the communication is used. The cover page 50 has a number of fields for designating sele