Apparatus and Method for Translating Facsimile Text Transmission

The invention generally relates to a facsimile translation machine that transmits or receives text images in a source natural language such as French or English. More particularly the invention relates to a facsimile machine that can recognize a source natural language within the text of a facsimile signal and translate the same source natural language to a desired target natural language and provide an output thereof.

The continuing popularity of facsimile machines as communication tools in the business community is evident from their worldwide availability and use. The industry standardization of a facsimile transmission protocol has permitted their global manufacture and use. However, despite the ability of a facsimile machine made and used overseas to communicate with an American-made machine, if the message sender and the message recipient are only able to express themselves in different natural languages, the facsimile machine alone cannot completely facilitate communication between them. As companies expand their international business activities the need to be able to traverse language barriers is clear.

A text translation machine that can be interfaced with a facsimile machine is described in U.S. Pat. No. 5,175,684 to Chong. Chong teaches translating a source natural language to a target natural language. However, Chong is incapable of recognizing the source language from the source natural language itself. A facsimile signal representative of a manually prepared, special cover sheet with predefined locations for providing specialized indicia designating the source language must precede facsimile transmission of the source natural language text to be translated. The Chong machine requires human intervention to prepare a special manual control signal to configure the machine for each transmission containing a different natural source language.

Not being able to recognize the source natural language of a facsimile transmission significantly limits the desirability of Chong. To use the translation capability of a remote machine of the type taught by Chong requires a sender using a standard facsimile machine to be aware of the capability, to have a copy of the special cover page, and to understand exactly how to identify the source natural language of her message. Given that the message sender and the message recipient in this situation need a translator to communicate, it may be difficult for the receiver to convey this information to the sender. Thus, Chong's requirement of a special manually prepared control signal to precede facsimile transmissions effectively creates a non-standard extension to the standard facsimile transmission protocol.

What is needed is a language recognition and translation facsimile machine that can automatically recognize the source language directly from the text images being received without having to resort to a non-standard prologue to identify the source natural language.

Apparatus for translating a facsimile signal representative of text in source natural language into a target natural language text signal has a facsimile input receiver, an optical character recognizer for converting the facsimile signal to a source natural language text signal, a source language recognizer, a translator for converting the source natural language to a target natural language signal, natural language recognition and a portion on another input device for producing a visually perceivable target natural language output.

The apparatus can signal a remote facsimile machine from which it is to receive a transmission to send facsimile signals representative of the source natural language text in a high resolution mode to aid in providing more accurate translations.

The facsimile input includes a facsimile machine or fax-modem for receiving an inbound facsimile signal representative of text images in a source natural language or a page scanner for scanning in hard copy images of text. The optical character recognizer includes an optical character recognition and conversion software module executable in a microprocessor for converting the input signal representative of a text image in a source natural language from the facsimile machine, fax-modem or page scanner into an ASCII text source natural language signal representative of the source natural language text. The natural language recognizer is executed by the microprocessor. The natural language recognizer acts upon the ASCII text source natural language signal in combination with data from a database of one or more natural languages. An etymologizer software module while being executed by the microprocessor causes the microprocessor to perform lookups for text matches in the first database to determine the source natural language of the ASCII text directly and exclusively from the ASCII text source natural language signal itself without the need for separately coded entries indicative of the source natural language.

The natural language translator includes the microprocessor that executes a software for translating the OCR output of the source natural language text image to a target natural language translation text. The natural language translator also has a first dictionary database for relating words of the source natural language to words of the target natural language. A lexicography software module is executed by the microprocessor to generate a second database containing undefined source natural language words by finding source natural language words present in the ASCII source natural language text signal but not contained in the first database.

The logical analysis and grammar correlation module creates an image of the sentence by analyzing interdependencies and establishing links between miscellaneous component words: subject, verb, object, subordinate clause, etc. Logical structures are then correlated against grammar constructs until a "legal" structure is arrived at, and then passed on to the target language generation program.

The logic and syntactic parser module has a dual goal--first, to build a model of the linguistic structure of the sentence and, second, in so doing, to eliminate syntactic ambiguities, i.e., to assign a syntactically accurate category to words with several possibilities. To carry out this analysis, the system uses recursive structural trees which are gradually built as the words are being parsed. Therefore, no structure is predetermined, illegal, or limited. The only common feature is recursivity, in that the internal structure of a phrase includes the anchorage point of a large number of similar or dissimilar phrases, all with an identical anchorage point. Thus, the network can continue to ramify as long as the words keep coming, and as long as they are logically interrelated. While building up the structure, the system uses three constituent levels: individual words which can be used separately, such as adverbs for example; clusters of words which make up a functional group in the sentence, such as noun-groups or verb-groups, with such functions as object, subject, etc.; and clauses which are made up of word clusters or individual words and which can themselves be related to other clauses. A sentence is a group of clauses. The tree-like structures are custom-built to reproduce the logical structure of the sentence in the source language, and to the inference engine used to build them. In particular, the inference engine is generalized to allow analysis of structures in various known languages, and to reject all incomplete or incorrect sentences which do not result in a logical structure. The inference engine uses a mathematical representation of language, is fully recursive and re-entrant, and self-determines and modifies its own strategy depending on the current structure, and incoming components in the sentence to be translated. It systematically applies the concepts of artificial intelligence to the field of linguistics to allow complete and systematic analysis of natural language.

The inventive apparatus also includes software for execution by the microprocessor that checks the accuracy of a translation.

From time to time a definition retrieval software module is executed by the microprocessor causing the microprocessor to send the second database of undefined source natural language words in the form of a facsimile encoded signal to a remote site for human and/or machine translation. In response to an outside request, it also can receive a facsimile transmission representative of entries for a third database relating the previously undefined source natural words of the second database to words of the target natural language. A merge software module is executed by the microprocessor to update the first database with the contents of the third database.

The output system comprises the computer data and control interface and software for converting text in the target natural language to an output signal for a facsimile machine, a modem or a printer. The facsimile machine, the printer or the modem generate the target natural language output of the text as respectively a facsimile signal or a modem signal or as printed hard copy.

It is a principal aspect of the present invention to provide apparatus for natural language recognition and translation that identifies the source language directly from the source natural text alone and provides a translation of the text.

It is a further aspect of the invention to provide apparatus that activates and configures its own translation capability based solely on the text of the transmitted or received message.

Other aspects of the present invention will become apparent to one of ordinary skill in the art upon review of the specification and claims in light of the accompanying drawings.

FIG. 1 is a perspective view of a translating facsimile machine embodying the present invention;

FIG. 2 is a schematic side elevational view of the translating facsimile transmission machine shown in FIG. 1 showing details of a paper handling train therein;

FIG. 3 is a schematic view of the translating facsimile machine shown in FIG. 1, showing details of an input paper handling path therein;

FIG. 4 is a schematic side view of the translating facsimile machine shown in FIG. 1, showing details of the paper paths together with a scanner head and a thermal printer;

FIG. 5 is a elevational view of a portion of the translating facsimile machine shown in FIG. 1 showing the location, in dotted form, of the thermal printer and the paper drive gear trains;

FIG. 6 is a partially schematic view of the electrical organization of the translating facsimile machine shown in FIG. 1;

FIG. 7 is a block diagram of electrical portions of the translating facsimile machine shown in FIG. 1;

FIG. 8 is a block diagram of an interface board portion of the electrical circuitry of the translating facsimile machine shown in FIG. 1;

FIG. 9 is a schematic diagram of a phone line simulator circuit for the translating facsimile machine shown in FIG. 1;

FIGS. 10a-b are a schematic diagram of a liquid crystal display interface circuit of the translating facsimile machine shown in FIG. 1;

FIGS. 11a-b are a schematic diagram of an ISA bus decoder of the translating facsimile machine shown in FIG. 1;

FIGS. 12a-b are a schematic diagram of a keyboard simulator/decoder circuit of the translating facsimile machine shown in FIG. 1;

FIG. 13 is a flow chart showing details of a keyboard handling software;

FIG. 14 is a flow chart showing additional details of the keyboard handling software;

FIG. 15 is a flow chart related to the operation of the translating facsimile machine shown in FIG. 1;

FIG. 16 is a flow chart of the operation of the translating facsimile machine shown in FIG. 1;

FIG. 17 is a flow chart of the operation of the translating facsimile machine shown in FIG. 1;

FIG. 18 is a flow chart of the translating steps performed by the translating facsimile machine shown in FIG. 1;

FIG. 19 is a flow chart of the steps taken by the translating facsimile machine and a remote facility to update the source-target language dictionary

FIG. 20 is a block diagram of an alternative embodiment of the inventive apparatus;

FIG. 21 is a block diagram of a second alternative embodiment of the inventive apparatus; and

FIG. 22 is a block diagram of a third alternative embodiment of the inventive apparatus.

Referring now to the drawings, and especially to FIGS. 1 and 6, apparatus for translating a facsimile signal representative of a source natural language to a target natural language, specifically a translating facsimile machine, is generally identified by numeral 10. The translating facsimile machine 10 includes a facsimile input receiver or jack 12, coupled to receive a facsimile signal and to send the facsimile input signal to an optical character recognizer 14 for converting the facsimile signal to a source natural language text signal. A source natural language recognizer 16 also receives the optical character recognition signal and determines whether the signal is representative of text in a recognizable source natural language. If the text of source natural language is recognized, a source natural language translator 18 converts the source natural language text signal to a target natural language text signal which may be stored in a storage unit or which may be output in visually perceptible form through an output device such as a printer 22. Alternatively, an additional output device 24 may be connected to the facsimile machine. Control of the facsimile machine is effected through a front panel 26.

As shown in FIG. 7, the apparatus 10 includes a Sharp Model UX184 facsimile machine, identified generally by numeral 28, having a panel board 30 that is electrically coupled through a panel board bus 32 to a keyboard decoder/simulator circuit 34. The keyboard decoder/simulator circuit 34 also is connected to a control circuit 36 in order to control various facsimile functions via a control bus 38.

The facsimile signals representative of the text of a source natural language are received by the jack 12 which supplies facsimile signal to a facsimile bus 40 connected to a phone line switcher simulator 42 connected via an internal facsimile bus 44 and a computer bus 46 to other portions of the system. The internal facsimile bus 44 is connected to the facsimile machine 28. A computer bus 46 is connected to a Zoom Telephonics 200M VFP fax-modem card 50. The fax-modem card 50 can communicate at 14,400 bps in conformance with the V.32 bis standard and runs under V. 42 bis/MNP for PC compatible systems. It is connected to a motherboard 51 via an ISA bus 52. The motherboard 51 is a Model PX-486 (SOJ Version) Half Size VL Bus board available from Micronics, Inc. of 232 E. Warren, Fremont, Calif. The motherboard 51 includes an Intel 486SX microprocessor, random access memory and read only memory. The motherboard 51 and ISA bus 52 comprises a portion of a computer 53. The ISA bus 52 is also connected to a disk controller unit 54 of the computer 53 having a serial port 56 and a parallel port 58. An IDE bus port 60 connects the disk controller unit 54 to a hard disk drive 62 and to a floppy disk drive 64. An ISA bus decoder 59 is connected to the ISA bus 52.

When a facsimile signal is received through the phone line 40, it is fed through the phone line switcher simulator 42 to the fax-modem card 50 which outputs a signal to the ISA bus 52.

Referring now to FIGS. 8 and 9, the phone line 40 is coupled through a J6 jack 210 which is a portion of the phone simulator 42. Signals may be exchanged from the phone line 40 coupled to the J6 jack 210 to a J4 jack 212 coupled to the facsimile line 44. The phone line simulator 42 may, in the alternative, steer signals between the J6 jack 210 and a J5 jack 214 which is connected to the two-line computer bus 46. The phone simulator 42 is coupled via an internal interface board data bus 216 to the ISA bus decoder 59 and to the keyboard simulator 34. Eight bits of data on the bus 216 are fed to a 74LS273 octal D flip-flop or latch 218 which is clocked by a signal on a relay line 220 extending from the ISA bus decoder 59 to the phone simulator 42. The data bits corresponding to bits D0 through D3, respectively, are output from Q outputs of the latch 218 to 74LS06inverters 222, 224, 226 and 228 to control a plurality of relays. The Q7 output drives an E.sub.-- LCD.sub.-- INT line 231 a. The Q8 output drives an E.sub.-- KBD.sub.-- INT line. The inverter 224 controls a double pole relay 230 having a pair of relay switches 232 and 234 and a relay solenoid 236. The signal from inverter 222 is fed over a line 240 to the solenoid 236. It should be appreciated that when inverter 222 switches low, current is allowed to flow through the coil 236 to the inverter energizing the coil and closing switches 232 and 234 which then allow a signal to be fed either to the computer bus 214 from the facsimile machine or from the computer back to the facsimile machine or from the computer 214 to the jack 210 connected to the phone line 40. Ultimate steering of the signals is controlled by the state of a relay 250 and a relay 252. The relay 250 is connected to a resistor network 254 via a pair of switches 256 and 258 controlled by a solenoid 260 coupled between a source of voltage an a control line 262. The potential on the control line 262 is controlled by the state of the inverter 224. The relay 252 includes switches 270 and 272 as well as a solenoid 274 connected via a line 276 to the inverter 228 to be controlled therefrom. Finally, a relay 280 comprising a pair of switches 282 and 284 and a solenoid 286 has the solenoid connected to a line 288 which is connected to the inverter 226. Thus, when the data bit from data line D2 is high and data bits D0, D1 and D3 are low, the relay 280 is closed connecting the facsimile bus 44 through the jack 212 to the phone line 40 via the jack 210. All of the other relays are left open. If it is desired to uncouple the facsimile bus directly from the phone line and to couple the computer to the phone line, the data bit D3 goes high, all other data bits are held low causing relay 270 to close while all other relays are allowed to be open, thereby connecting the J6 jack 210 directly to the J5 jack 214 which is coupled to the computer bus 46. If it is desired to output signals from the computer through the facsimile machine, the two highest order bits D2 and D3 are set low, lower to bits D1 and D0 are set high causing the relays 270 and 280 to be open and the relays 230 and 250 to be closed, coupling the computer to the facsimile lines through the voltage biasing network 254.

The liquid crystal display interface as is shown in detail in FIGS. 10a-b is coupled via a J7 jack 300 to a bus 301 connected to a liquid crystal display positioned on the front panel of the facsimile machine housing. Driver output signals are supplied through jack J7 and bus 301. The bus 301 drives a display 59. Data is received on the data bus 216 which is connected to a 74LS273octal-D flip-flop or latch 303 and to a 74LS245octal bus transceiver 304 controlled via a mode line 305 connected to its G pin. The transceiver receives outputs both from a 22V10 generic array logic device 306 (U6) and a 22V10 generic array logic device 307 (U9). Generic array logic devices 306 and 307 are programmed as set forth in the appendix for the respective portions indicated by numbers U6 and U9. Both generic array logic devices 306 and 307 receive inputs from the latch 303 tapped from the bus 216. It may be appreciate that data from the data bus 216 is also delivered to the J7 jack 300 for driving the LCD display. Interconnection is effected with the panel board or panel unit 30 via a J3A connector 309 and to the control board 36 via a J2A connector 310 to allow communication to travel back and forth between the liquid crystal display interface and the facsimile machine 28. In addition, the generic array logic device 306 drives an LCD.sub.-- INT line 312, a KBD.sub.-- INT line 314 as well as E.sub.-- LCD.sub.-- INT line 231a and E.sub.-- KBD.sub.-- INT line 231b.

An LCD.sub.-- INT line 317 is coupled between the LCD interface 59a and the ISA bus decoder. A KBD.sub.-- INTRQ line 318 is connected between the keyboard simulator/decoder 34 and the LCD interface 59a. An LCD/line 319 is coupled between the LCD interface 59a and the ISA bus decoder 59.

Translation of facsimile standard signals to ISA bus standard signals is achieved through the action of the keyboard simulator/decoder 34 and the ISA bus decoder 59.

As is shown in FIGS. 8 and 11a-b, the keyboard simulator/decoder communicates over an oscillator line 320, an E.sub.-- MS line 321, a SIM/line 322, a decode line 323, the KBD.sub.-- INT line 314, the KBD.sub.-- INTRQ line 318 and the data bus 216 with the ISA bus decoder 59, the LCD interface 59a and the phone line simulator 42.

Referring now in particular to FIGS. 11a-b, a J1 jack 324 having standard ISA bus pins is coupled via an internal decoder bus 325 to a GAL 22V12 generic array device 330 (U1) and a 74LS245 octal bus transceiver 326. The data bus 216 is coupled bidirectionally through the transceiver 326 via a plurality of lines 327 to the J1 jack 324. The generic array logic device 330 (U1) and 332 (U2) are internally programmed as set forth in the attached appendix. The generic array logic devices convert signals from the ISA bus standard to internal or facsimile media standard signals and vice versa. The ISA bus decoder 59 thus allows personal computer compatible signals of the ISA bus type to be converted to facsimile compatible signals of the type carried inside the facsimile machine as fed to and from the keyboard simulator/decoder 34.

The keyboard simulator/decoder 34 is shown in greater detail in FIGS. 12a-b. The bus 32 is connected to a J3 jack 390 and the control PWB unit bus 38 is connected to a J2 jack 392. The signals represented at jacks 390 and 392 are standard signals for the Sharp UX184 facsimile machine. A portion of the signals from the J2 jack are fed to the control unit in the facsimile machine via a GAL 22V10 generic array logic device 394 (U10), the encoding for which is set forth in the attached appendix.

The data bus 216 is in communication with a 22V10 generic array logic device 400 (U11) and a 74LS273 octal-D flip-flop or latch 402. The Q1 through Q11 outputs from latch 402 are fed to a 22V10 generic array logic device 404 (U6) which is programmed as set forth in the appendix and which is, in turn, connected the J2 jack 392 extending to the panel unit bus 32. Data lines D4 through D7 are also coupled to the generic logic device 400 which drives a portion of the J3 pins 390 connected to the panel unit bus 32.

Referring now to FIG. 13, a keyboard interrupt routine which is executed by the microprocessor 76 and is normally stored on the hard drive 62 is shown therein. The keyboard interrupt routine is stored in RAM 70 and in an step 500 awaits a keyboard interrupt signal. Upon receiving the keyboard interrupt signal, it reads a keyboard scan code in a step 502 and a keyboard sense code in a step 504. In a step 506, the keyboard scan and sense codes are combined in a single byte to form a keyboard key code which is decoded in a step 508. The routine keyboard interrupt is completed in a step 510. Key selection is determined in a key selection routine shown in FIG. 14. The key selection routine is initiated in a step 520 that determines the proper key depression from a keyboard state table. A key state byte is written in the step 522. In a step 524 a tenth of a second delay simulates a holding key depression and in a step 526 a KEYBM-SIM variable is cleared. The routine is exited in a step 528.

When the apparatus 10 receives a facsimile signal representative of source language text, as shown in FIG. 15, a low level routine is executed at a step 540 to uncouple the microprocessor 76 from the phone line, set the facsimile 28 to receive the phone signal, enable the facsimile LCD interrupt and enable a keyboard interrupt. In a step 542, all facsimile key selections are echoed to the facsimile machine 28, with the exception of specialty keys. In a step 554, the facsimile LCD activity is monitored by storing LCD data in controls and decoding data to determine when the transmission has been completed. In a step 546, when the facsimile transmission has been completed, the facsimile is removed from the phone line and control is transferred back to the computer to receive phone line signals. The LCD interrupt is disabled and the keyboard interrupt is enabled. The routine is exited in a step 548.

When it is desired to perform translation, as will be seen hereinafter, a low level routine as set forth in FIG. 16 is enabled. In a step 570, the computer is uncoupled from the phone line, the computer is coupled to the facsimile machine 28, the facsimile machine 28 is uncoupled from the phone line and the facsimile signal is sent to the computer. The LCD interrupt is disabled and the keyboard interrupts are enabled. Optionally, in a step 572, the LCD may set the computer to write the facsimile LCD. A test is performed in a step 574 to determine whether the facsimile is in manual mode. If it is not, control is transferred to a step 576 to simulate depressing a receive mode key. If it is in manual mode, the read facsimile mode is enabled and a document facsimile may be sent with the computer board in a step 578 by transferring control to a step 580 to simulate pressing a start key on the facsimile. The transmission is awaited to be completed in a step 582 and, when completed, transfers control to a step 584, uncoupling the computer from the facsimile and removing the facsimile from the computer, coupling the computer with the phone line, disabling the facsimile LCD interrupt and enabling the keyboard interrupt. The routine is exited in a step 586.

More specifically, when it is desired to receive a facsimile transmission signal at a higher level, a routine is executed in the random access memory with an initiation step 600, followed by a decision step 602 to determine whether input has been detected. If it has, control is transferred to a decision step 604 to determine whether the input is from a hard copy being scanned by the thermal scanner or from a modem. If the input comes through the modem, control is transferred through a step 606 to disable the facsimile machine and scanner and to generate an acknowledge modem signal in step 608. In the step 610 ASCII text is read in from data fields and stored in the hard drive in the step 612. In the event that a facsimile signal representative of the source natural language text is received, control is transferred from step 604 to a step 620 to disable the scanner and modem and a step 622 to acknowledge the facsimile machine. During the step 622 the apparatus 10 may signal the sending system that contacts it to send its image transmission in high resolution or "fine" mode. This would allow the apparatus 10 to translate more accurately than from text images received at normal resolution. Data in the form of a facsimile signal is read in in a step 624 and converted to bit mapped data signal in a step 626. The bit mapped data signal is converted by an optical character recognition program, such as Winfax Pro or optical character recognition software of the type that may be purchased from Caere Corporation, to an ASCII text signal of the source natural language text in a step 628 and then passed to the step 612.

If hard copy is being scanned, control is transferred to a step 640 to disable the facsimile machine and modem. The page scanner output is acknowledged at a step 642 and bit mapped data is read in a step 644. Control is then transferred to the step 628 to convert the bit mapped data to ASCII text which is stored on a hard drive in step 612. In a step 650, following step 612, a determination is made as to the source natural language, for instance to determine whether the language is in English or French. If the source natural language is different than the target language as determined by a decision in the step 652, a translation engine is started in a step 654, the details of which will be discussed hereafter. If the source natural language is identical to the target natural language, control is transferred to a decision block 656 to determine whether output is desired through a modem in the step 658 to a translated facsimile in a step 660 or to the printer in a step 662.

Referring now to FIG. 18, a translation routine 654 is generally shown therein and begins with a step 700 in which the next word in the source target dictionary is looked up. If a definition is present, as determined by a comparison step 602, control is transferred to a step 704 to store word and possible definitions in a temporary glossary. If the definition is not present, the words are stored in an undefined text area on the disk in a step 706.

The splitter module divides the text into paragraphs, sentences and separate words. Words are classified in four categories: alphabetic, figures and numbers, punctuation, and hybrids. The lemmatizer or dictionary access module starts from words defined by the previous module and establishes a list of possible translations for each word, along with compound expressions, if any.

The semantic disambiguation module determines optimal translations and eliminates those which have no relevance in the context. The system keeps track of related contextual words use in the overall text, the paragraphs, and the sentence currently being analyzed. Context words receive a weight proportional to the inverse square of their distance to the ambiguous word. Such disambiguation is not dependent upon a binary decision, but rather a selection based on a large number of possible levels, constantly and gradually modified from a sentence to the next, as text flows on. Programming is based on the systematic application of the principles of fuzzy multilevel or two valued logic in order to more closely approximate the linguistic behavior of the human mind.

Control is passed through another decision block 710 to determine whether more words need to be looked up.

If the answer is "yes", control is transferred back to the step 700. If not, control is transferred to a step 712 to create the possible translations via a lexical tree generator. Once the lexical tree generator has generated a multi-dimensional tree structure related to the possible language translation, control is passed to a step 714 to analyze the next possible translation from a different tree structure for the best fit. A comparison step is made in step 716 to determine whether to eliminate one or the other trees. If not, control is transferred to a step 718 to add to a sentence image. If it is desired to eliminate the translation, control is transferred to a step 720, removing that translation. A test is made in the step 722 to determine whether more possible translations should be made. If so, control is transferred back to step 712.

If not, control is transferred to a step 730 to populate a sentence image following which a raw target translation is generated in the step 732. The sentences are polished in a step 734 and then may be output through an appropriate output means. The target language generation module walks through the structured map of the sentence and the module regenerates the sentence in the target language, while applying all grammatical rules (particularly agreement rules in the romance languages). The editing and polishing module takes care of a number of cosmetic operations, such as elision, contractions, etc. At the output of this last module, the target natural language translation is available, and can be forwarded to a monitor or saved to a file for further editing. This file normally includes the text in the source natural language, plus the translated or target natural language text, paragraph by paragraph in order to facilitate post-editing and correction. The overall process is similar to the mechanisms used by a conference interpreter in a simultaneous translation situation.

As an additional feature the apparatus 10 may take words collected in the database of unidentifiable words and automatically dial a remote site service provider as is provided in step 760 shown in FIG. 19. The words either would be downloaded in facsimile or binary format to the remote service in a step 762. The service provider would then either machine translate or translate manually the unknown words and assign meanings to them in a step 764. The remote site would then dial up the apparatus 10 in a step 766. In a step 768 the newly translated words and their associated target language meanings would then sent by facsimile or binary transmission back to the originating apparatus 10. In a step 770, the originating apparatus 10 would then merge then newly translated words and their meanings into the original database containing the target language dictionary for later use. Such an automated feature would enable to learn automatically and improve its accuracy of translation with experience.

As is shown in FIG. 20, a system can also be constructed wherein generally the computer 53 may be connected via an RS-232 serial line 900 into a facsimile machine 902 having an RS-232 serial port to communicate therewith. The facsimile machine also includes a modem 904 which is connectable to a phone line 906 with which it communicates.

As is shown in FIG. 21 in a second alternative embodiment, the computer 53 is incorporated in a pedestal unit 910 having an interface board 912 coupled between the computer 53 and a modem 914. A switched logic bus 916 is coupled to switching logic 920 which allows communication of the modem 914 via a line 922 either with a phone line 924 or a line 926. A facsimile machine 930 having modem 932 is coupled to the line 926.

In a third alternative embodiment, as is shown in FIG. 22, a pedestal unit 930, including the computer 53, has an RS-232 serial communications system 932 connected via a line 934 with an RS-232 serial communication system 936 of a facsimile machine 938. An interface board 940 couples a modem 942 to the computer 53. Switching logic 944 sends and receives signals, including control signals, via a bus 946 to the interface board 940. In order to control whether modem 942, which is connected via line 950 to the switched logic 944, is fed through the switch logic to the phone line 960 or to a modem line 962 coupled to a modem 964 of the facsimile machine 938.

These three alternative embodiments thus allow a substantially stand-alone computer unit to be connected either by an RS-232 serial communications line with a facsimile machine or via a modem or both.

While there have been illustrated and described a particular embodiment of the present invention, it will be appreciated that numerous changes and modifications will occur to those skilled in the art, and it is intended in the appended claims to cover all those changes and modifications which fall within the true spirit and scope of the present invention. ##SPC1##