The innovation is now described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding thereof. It may be evident, however, that the innovation can be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form in order to facilitate a description thereof.
As used in this application, the terms “component” and “system” are intended to refer to a computer-related entity, either hardware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to being, a process running on a processor, a processor, a hard disk drive, multiple storage drives (of optical and/or magnetic storage medium), an object, an executable, a thread of execution, a program, and/or a computer. By way of illustration, both an application running on a server and the server can be a component. One or more components can reside within a process and/or thread of execution, and a component can be localized on one computer and/or distributed between two or more computers.
The disclosed architecture employs machine translation technology, at least in terms of application development, to automatically translate a single developed dataset into a full set of desired target languages. Machine translation automates the otherwise costly and time-consuming process of translating from one human language to another. This reduces the effort required to develop multiple training and test sets, one for each target language, into the effort required to develop datasets in a single language. The disclosed architecture facilitates functional testing of the underlying natural language technology being developed across all target languages, exposing any language-specific idiosyncrasies that may exist. Although described in the context of natural language processing (NLP), the disclosed architecture also finds application in automatic speech recognition (ASR) systems and text translation systems.
Referring initially to the drawings,
It is to be understood that the disclosed machine translation architecture can include and/or access components that facilitate or provide some or all of at least the following example data and processes that facilitate understanding humans via natural language processing and/or speech recognition: information retrieval, extraction and inferencing related to phonetics and phonology (how words are pronounced in colloquial speech), parsing, morphological analysis (about the shape and behavior of words in context), lexical semantics (the meanings of the component words), lexical ambiguity, syntactical analysis (about the ordering and grouping of words), pragmatics (use of polite and indirect language), language dictionaries, statistical rules, linguistic rules, lexical lookup methods, semantics processing, compositional semantics (knowledge of the how component words combine to form larger meanings), speech segmentation, text segmentation, word sense disambiguation, contextual processing, temporal and/or spatial reasoning, speech acts or plans (for dealing with sentences or phrases that do not mean what is literally expressed), discourse conventions, and imperfect or irregular input (for dealing with foreign or regional accents, vocal impediments, and typing or grammatical errors). Moreover, it is within contemplation of the subject architecture that statistical natural language processing can be utilized that employs stochastic, probabilistic and statistical methods to resolve some of the more complex processes referred to above, as well as pattern-based machine translation technologies.
Additionally, the machine translation component 104 is not limited by the type of translation engine, and thus, can utilize engines that are based on a direct (or transformer) architectures, or indirect (or linguistic knowledge) architectures, for example.
At 200, an authoring tool is received that is utilized for application development. The authoring tool can be a standalone program that allows a user to write program code. Alternatively, the authoring tool can be a considered a suite of programs as associated with an integrated development environment and/or an application development environment that includes a set of programs which can be run from a single user interface, such as a programming language that also includes a text editor, compiler and debugger, for example. In one example implementation, the authoring tool user interface facilitates use of a grammar builder program via which the author can describe responses to prompts which the application being developed is expected to receive and process. The responses can be presented by a user as utterances and/or text inputs. At 202, a first dataset of natural language training data is generated in a first human language. At 204, the first dataset is machine translated into a second natural language dataset of a different human language. At 206, the second dataset is tested at least for performance. If the tested dataset successfully meets the desired test criteria, the second dataset is employed in the application being developed, as indicated at 208.
Referring now to
At 302, a top level concept (or rule) is defined and associated with a response container. Here, the author can describe responses to a prompt which the application is expected to handle. The author (or application developer) typically defines the top level rule to be associated with a particular dialog element, or “question answer,” in the application.
A response container can contain one or more response nodes, which response nodes define the individual high level concepts that are handled by the application. Accordingly, at 304, response concepts are defined for underlying response nodes of the tree. For example, consider a retail application example having a top level rule of “How May I Help You?” The response container could hold the following five response nodes: 1) “Get Store Hours”, 2) “Locate Nearest Store”, 3) “Get Driving Directions”, 4) “Check Inventory Availability”, and 5) “Order Status Inquiry”.
At 306, after defining the response nodes within the response container, the developer populates each of the nodes with a collection of example sentences (or utterances) that represent the many ways a user interacting with the system could articulate the concept being conveyed. For example, the “Get Store Hours” node can contain utterances similar to “How late are you open today?”, “What are your store hours?”, “What time do you open?”, “Are you open on Sunday?”, and so on.
After each of the response containers and their underlying response nodes have been fully defined, that is, when all of the response nodes for each response container defined in the application have been populated with all of the example utterances the developer wishes to include, the developer can initiate machine translation of the container(s) and associated nodes (e.g., example utterances) to output a natural language dataset in a different human language, as indicated at 308.
In another implementation, the machine translation process facilitates output of multiple natural language datasets each in its own human language.
At 310, testing can be performed on one or more of the output datasets in accordance with predetermined testing criteria. The criteria can be employed to provide a success or failure indication as to the quality of the output dataset in processing test data. In another implementation, metrics are employed that indicate a degree of success or failure, thereby providing a more accurate representation of the quality of the dataset. If successful, the language dataset can be employed in the desired application, as indicated at 312.
In one implementation, the input dataset 402 is intended to be a “master” dataset from which all other output datasets will be created by machine translation. In another implementation, it is to be understood that the dataset 402 can represent multiple different input datasets each of which includes training data, and optionally, test data, and from which the desired output datasets are generated. For example, it is to be appreciated that a first dataset may, over time, prove to be a better “fit” for machine translation into the many dialects of the Chinese language, rather than a second input dataset, which proves to be a better “fit” for Middle Eastern dialects. Accordingly, these different input datasets can be stored and automatically retrieved based on the desired output languages. Thereafter, machine translation can be utilized to more effectively provide the desired output natural language datasets.
As indicated supra, the developer can manually enter information, expressions, etc., into the input dataset 402. Alternatively, or in combination therewith, an import component 410 facilitates importing the desired information, expressions, utterances, etc., into the system 400 from other files and/or file formats, for more expedient development. This capability significantly reduces the time the developer would need to take to re-enter the information manually into the response containers and response nodes, for example. The import component 410 can be a software capability provided as program menu option for importing (or exporting) files and/or other types of data, which capability can be commonly found in conventional software applications. Alternatively, a separate program can be provided that receives incompatible formats (e.g., proprietary formats) and converts this information into a format suitable for importation and processing by the authoring tool.
The system 400 can employ a language selection component 412 that interfaces the machine translation component 104 to a language component 414 for selecting one or more human languages 416 (denoted HL1, . . . , HLM, where M is a positive integer) into which the input dataset 402 will be translated. The languages 416 can be in the form of language models that can be readily updated as needed. Selection of the languages 416 can be via a menuing system of a user interface, for example.
Once the languages 416 are selected, the machine translation component 104 translates the completed input dataset(s) 402 into the corresponding output human language datasets 404 (denoted in this example as three datasets HLDS1, HLDS3, and HLDS10 that correspond to three selected human languages HL1, HL3, and HL10 of the language component 414).
A replacement component 418 facilitates insertion of the machine translated natural language expressions (or data) back into the corresponding locations of the response container tree(s) to arrive at the final output natural language dataset.
A tagging component 420 facilitates tagging of selected training data 406 for generating the test data 408. Although represented as a block separate from the training data 406, the test data 408 represents training data that has been automatically selected and grouped for testing purposes. As a separate block, the test data 408 can be a copy of the tagged training data which is then set aside for testing and analysis purposes.
Although the machine translation engine and related components have been described in combination with a development tool, it is to be understood that the engine/components can be a standalone application that interfaces to the tool 400 to provide the disclosed functionality.
The developer can be selective about which information to translate in a container tree. In other words, it is not a requirement that the whole container tree be translated. For example, translation via the machine translation component 104 can be performed at the response node level by selecting one or more of the response nodes 702, for example, the first response node RN1 and associated example utterances 704. Response node level translation can be performed by selecting a machine translation function for the desired node, followed by selecting the desired target language(s). In one implementation, selection of the desired target language automatically triggers the machine translation process for the entire tree(s) or just the nodes.
Alternatively, selection of the first response container RC1 can trigger the machine translation process for all of the example utterances (704, 706 and 708) in the corresponding response nodes 702 contained therein. The individual example utterances can then be replaced by their machine translated substitutes.
Thereafter, the authoring tool can utilize these translated examples as an input to train models for ASR systems and/or NLP systems, for example. Additionally, as indicated herein, one or more example utterances within a response node can be tagged as being slated for testing purposes, which enables the use of the disclosed novel technology for developing both training and testing data for the desired systems.
A classifier is a function that maps an input attribute vector, x=(x1, x2, x3, x4, xn), to a class label class(x). The classifier can also output a confidence that the input belongs to a class, that is, f(x)=confidence(class(x)). Such classification can employ a probabilistic and/or other statistical analysis (e.g., one factoring into the analysis utilities and costs to maximize the expected value to one or more people) to prognose or infer an action that a user desires to be automatically performed.
As used herein, terms “to infer” and “inference” refer generally to the process of reasoning about or inferring states of the system, environment, and/or user from a set of observations as captured via events and/or data. Inference can be employed to identify a specific context or action, or can generate a probability distribution over states, for example. The inference can be probabilistic—that is, the computation of a probability distribution over states of interest based on a consideration of data and events. Inference can also refer to techniques employed for composing higher-level events from a set of events and/or data. Such inference results in the construction of new events or actions from a set of observed events and/or stored event data, whether or not the events are correlated in close temporal proximity, and whether the events and data come from one or several event and data sources.
A support vector machine (SVM) is an example of a classifier that can be employed. The SVM operates by finding a hypersurface in the space of possible inputs that splits the triggering input events from the non-triggering events in an optimal way. Intuitively, this makes the classification correct for testing data that is near, but not identical to training data. Other directed and undirected model classification approaches include, for example, naïve Bayes, Bayesian networks, decision trees, neural networks, fuzzy logic models, and probabilistic classification models providing different patterns of independence can be employed. Classification as used herein also is inclusive of statistical regression that is utilized to develop models of ranking or priority.
As will be readily appreciated from the subject specification, the subject invention can employ classifiers that are explicitly trained (e.g., via a generic training data) as well as implicitly trained (e.g., via observing user behavior, receiving extrinsic information). For example, SVM's are configured via a learning or training phase within a classifier constructor and feature selection module. Thus, the classifier(s) can be employed to automatically learn and perform a number of functions according to predetermined criteria.
In one implementation, the MLR component 1002 can learn and reason about which of multiple input datasets to use for translation processing. For example, as indicated supra, the developer can define many different datasets over time, some of which operate to translate better for the desired output languages. In operation, when the developer selects the output language(s), the MLR component 1002 can recommend that a specific input dataset be employed, since, as learned in the past, this dataset shows a higher rate of success for translation than another. Although the disclosed architecture describes use of a single input dataset for translation into the many output languages, it is to be appreciated that based on testing, an input dataset can be computed to be less than optimal for translation into the desired output languages. However, this dataset may prove to be a better dataset for translation into other languages than currently desired. Accordingly, the developer can save these many different versions of input datasets for later use. Based on this swapping in and out of input datasets to arrive at the optimal output languages, the MLR component 1002 can learn and reason about this, thereafter recommending one input dataset over another, for example, based on the desired output languages.
In another implementation, the MLR component 1002 can perform cost/benefit analysis based on the type of machine translation engine utilized for the input dataset and desired output dataset languages, and therefrom, suggest that another type of engine may provide an improvement on the translation process.
In yet another implementation, this type of translation management can be reduced to a lower level, wherein the MLR component 1002 operates to learn and reason about which of the data (at the node level, for example) in the training dataset to tag for utilization as the testing dataset.
These are only but a few examples of the flexibility that can be employed by the MLR component 1002, and are not to be construed as limiting in any way. For example, in still another implementation, learning and reasoning can be applied to determining the number and type of example utterances to generate for a given response node, the number of containers for the application, and so on. The number of example utterances required for translation into a Chinese dialect may be fewer than the number required for translation into English, for example.
Accordingly, the MLR component facilitates at least maintaining a system according to the desired metrics. Moreover, it can be appreciated that in many cases, the system can be improved upon based on changes that occur in the underlying data, and other system parameters.
At 1200, development of a natural language training dataset is initiated. At 1202, entry of the response container and associated example utterances for the response nodes, is completed, in preparation for translation. At 1204, the developer selects the first language for machine translation. The system can then check if the first selected language is normally associated with a multi-lingual populace and/or if the application being developed is slated for use in an area of multi-lingual users, as indicated at 1206. If so, at 1208, the developer can then manually select a second language in which the populace is normally fluent for that area. Alternatively, the system presents lists of languages from which to select the most likely second language for this dataset. At 1210, the system machine translates both the first and second languages for the concept tree(s), and inserts the translated data back into the tree(s) at the appropriate places. Thus, a single example utterance will be replaced with two translated utterances: one in the first language, and the other in the second language. If is determined not to be a multilingual populace, flow is from 1206 to 1212, to machine translate as would be performed normally.
Referring now to
Generally, program modules include routines, programs, components, data structures, etc., that perform particular tasks or implement particular abstract data types. Moreover, those skilled in the art will appreciate that the inventive methods can be practiced with other computer system configurations, including single-processor or multiprocessor computer systems, minicomputers, mainframe computers, as well as personal computers, hand-held computing devices, microprocessor-based or programmable consumer electronics, and the like, each of which can be operatively coupled to one or more associated devices.
The illustrated aspects of the innovation may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote memory storage devices.
A computer typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by the computer and includes both volatile and non-volatile media, removable and non-removable media. By way of example, and not limitation, computer-readable media can comprise computer storage media and communication media. Computer storage media includes both volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by the computer.
With reference again to
The system bus 1408 can be any of several types of bus structure that may further interconnect to a memory bus (with or without a memory controller), a peripheral bus, and a local bus using any of a variety of commercially available bus architectures. The system memory 1406 includes read-only memory (ROM) 1410 and random access memory (RAM) 1412. A basic input/output system (BIOS) is stored in a non-volatile memory 1410 such as ROM, EPROM, EEPROM, which BIOS contains the basic routines that help to transfer information between elements within the computer 1402, such as during start-up. The RAM 1412 can also include a high-speed RAM such as static RAM for caching data.
The computer 1402 further includes an internal hard disk drive (HDD) 1414 (e.g., EIDE, SATA) on which the various authoring tool and machine translation components can be stored, which internal hard disk drive 1414 may also be configured for external use in a suitable chassis (not shown), a magnetic floppy disk drive (FDD) 1416, (e.g., to read from or write to a removable diskette 1418) and an optical disk drive 1420, (e.g., reading a CD-ROM disk 1422 or, to read from or write to other high capacity optical media such as the DVD). The hard disk drive 1414, magnetic disk drive 1416 and optical disk drive 1420 can be connected to the system bus 1408 by a hard disk drive interface 1424, a magnetic disk drive interface 1426 and an optical drive interface 1428, respectively. The interface 1424 for external drive implementations includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies. Other external drive connection technologies are within contemplation of the subject innovation.
The drives and their associated computer-readable media provide nonvolatile storage of data, data structures, computer-executable instructions, and so forth. For the computer 1402, the drives and media accommodate the storage of any data in a suitable digital format. Although the description of computer-readable media above refers to a HDD, a removable magnetic diskette, and a removable optical media such as a CD or DVD, it should be appreciated by those skilled in the art that other types of media which are readable by a computer, such as zip drives, magnetic cassettes, flash memory cards, cartridges, and the like, may also be used in the exemplary operating environment, and further, that any such media may contain computer-executable instructions for performing the methods of the disclosed innovation.
A number of program modules can be stored in the drives and RAM 1412, including an operating system 1430, one or more application programs 1432 (e.g., the authoring tool, machine translation engine, . . . ), other program modules 1434 and program data 1436. All or portions of the operating system, applications, modules, and/or data can also be cached in the RAM 1412. It is to be appreciated that the innovation can be implemented with various commercially available operating systems or combinations of operating systems.
A user can enter commands and information into the computer 1402 through one or more wired/wireless input devices, for example, a keyboard 1438 and a pointing device, such as a mouse 1440. Other input devices (not shown) may include a microphone, an IR remote control, a joystick, a game pad, a stylus pen, touch screen, or the like. These and other input devices are often connected to the processing unit 1404 through an input device interface 1442 that is coupled to the system bus 1408, but can be connected by other interfaces, such as a parallel port, an IEEE 1394 serial port, a game port, a USB port, an IR interface, etc.
A monitor 1444 or other type of display device is also connected to the system bus 1408 via an interface, such as a video adapter 1446. In addition to the monitor 1444, a computer typically includes other peripheral output devices (not shown), such as speakers, printers, etc.
The computer 1402 may operate in a networked environment using logical connections via wired and/or wireless communications to one or more remote computers, such as a remote computer(s) 1448. The remote computer(s) 1448 can be a workstation, a server computer, a router, a personal computer, portable computer, microprocessor-based entertainment appliance, a peer device or other common network node, and typically includes many or all of the elements described relative to the computer 1402, although, for purposes of brevity, only a memory/storage device 1450 is illustrated. The logical connections depicted include wired/wireless connectivity to a local area network (LAN) 1452 and/or larger networks, for example, a wide area network (WAN) 1454. Such LAN and WAN networking environments are commonplace in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which may connect to a global communications network, for example, the Internet.
When used in a LAN networking environment, the computer 1402 is connected to the local network 1452 through a wired and/or wireless communication network interface or adapter 1456. The adaptor 1456 may facilitate wired or wireless communication to the LAN 1452, which may also include a wireless access point disposed thereon for communicating with the wireless adaptor 1456.
When used in a WAN networking environment, the computer 1402 can include a modem 1458, or is connected to a communications server on the WAN 1454, or has other means for establishing communications over the WAN 1454, such as by way of the Internet. The modem 1458, which can be internal or external and a wired or wireless device, is connected to the system bus 1408 via the serial port interface 1442. In a networked environment, program modules depicted relative to the computer 1402, or portions thereof, can be stored in the remote memory/storage device 1450. It will be appreciated that the network connections shown are exemplary and other means of establishing a communications link between the computers can be used.
The computer 1402 is operable to communicate with any wireless devices or entities operatively disposed in wireless communication, for example, a printer, scanner, desktop and/or portable computer, portable data assistant, communications satellite, any piece of equipment or location associated with a wirelessly detectable tag (e.g., a kiosk, news stand, restroom), and telephone. This includes at least Wi-Fi and Bluetooth™ wireless technologies. Thus, the communication can be a predefined structure as with a conventional network or simply an ad hoc communication between at least two devices.
Referring now to
The system 1500 also includes one or more server(s) 1504. The server(s) 1504 can also be hardware and/or software (e.g., threads, processes, computing devices). The servers 1504 can house threads to perform transformations by employing the invention, for example. One possible communication between a client 1502 and a server 1504 can be in the form of a data packet adapted to be transmitted between two or more computer processes. The data packet may include a cookie and/or associated contextual information, for example. The system 1500 includes a communication framework 1506 (e.g., a global communication network such as the Internet) that can be employed to facilitate communications between the client(s) 1502 and the server(s) 1504.
Communications can be facilitated via a wired (including optical fiber) and/or wireless technology. The client(s) 1502 are operatively connected to one or more client data store(s) 1508 that can be employed to store information local to the client(s) 1502 (e.g., cookie(s) and/or associated contextual information). Similarly, the server(s) 1504 are operatively connected to one or more server data store(s) 1510 that can be employed to store information local to the servers 1504.
What has been described above includes examples of the disclosed innovation. It is, of course, not possible to describe every conceivable combination of components and/or methodologies, but one of ordinary skill in the art may recognize that many further combinations and permutations are possible. Accordingly, the innovation is intended to embrace all such alterations, modifications and variations that fall within the spirit and scope of the appended claims. Furthermore, to the extent that the term “includes” is used in either the detailed description or the claims, such term is intended to be inclusive in a manner similar to the term “comprising” as “comprising” is interpreted when employed as a transitional word in a claim.