Showing posts with label Lili Cheng. Show all posts
Showing posts with label Lili Cheng. Show all posts

Artificial Intelligence - Machine Translation.


Machine translation is the process of using computer technology to automatically translate human languages.

The US administration saw machine translation as a valuable instrument in diplomatic attempts to restrict communism in the USSR and the People's Republic of China from the 1950s through the 1970s.

Machine translation has lately become a tool for marketing goods and services in countries where they would otherwise be unavailable due to language limitations, as well as a standalone offering.

Machine translation is also one of the litmus tests for artificial intelligence progress.

This artificial intelligence study advances along three broad paradigms.

Rule-based expert systems and statistical methods to machine translation are the earliest.

Neural-based machine translation and example-based machine translation are two more contemporary paradigms (or translation by analogy).

Within computer linguistics, automated language translation is now regarded an academic specialization.

While there are multiple possible roots for the present discipline of machine translation, the notion of automated translation as an academic topic derives from a 1947 communication between crystallographer Andrew D. Booth of Birkbeck College (London) and Warren Weaver of the Rockefeller Foundation.

"I have a manuscript in front of me that is written in Russian, but I am going to assume that it is truly written in English and that it has been coded in some bizarre symbols," Weaver said in a preserved note to colleagues in 1949.

To access the information contained in the text, all I have to do is peel away the code" (Warren Weaver, as cited in Arnold et al. 1994, 13).

Most commercial machine translation systems have a translation engine at their core.

The user's sentences are parsed several times by translation engines, each time applying algorithmic rules to transform the source sentence into the desired target language.

There are rules for word-based and phrase-based trans formation.

The initial objective of a parser software is generally to replace words using a two-language dictionary.

Additional processing rounds of the phrases use comparative grammatical rules that consider sentence structure, verb form, and suffixes.

The intelligibility and accuracy of translation engines are measured.

Machine translation isn't perfect.

Poor grammar in the source text, lexical and structural differences between languages, ambiguous usage, multiple meanings of words and idioms, and local variations in usage can all lead to "word salad" translations.

In 1959–60, MIT philosopher, linguist, and mathematician Yehoshua Bar-Hillel issued the harshest early criticism of machine translation of language.

In principle, according to Bar-Hillel, near-perfect machine translation is impossible.

He used the following sentence to demonstrate the issue: John was on the prowl for his toy box.

He eventually discovered it.

In the pen, there was a box.

John was overjoyed.

The word "pen" poses a problem in this statement since it might refer to a child's playpen or a writing ballpoint pen.

Knowing the difference necessitates a broad understanding of the world, which a computer lacks.

When the National Academy of Sciences Automatic Language Processing Advisory Committee (ALPAC) released an extremely damaging report about the poor quality and high cost of machine translation in 1964, the initial rounds of US government funding eroded.

ALPAC came to the conclusion that the country already had an abundant supply of human translators capable of producing significantly greater translations.

Many machine translation experts slammed the ALPAC report, pointing to machine efficiency in the preparation of first drafts and the successful rollout of a few machine translation systems.

In the 1960s and 1970s, there were only a few machine translation research groups.

The TAUM group in Canada, the Mel'cuk and Apresian groups in the Soviet Union, the GETA group in France, and the German Saarbrücken SUSY group were among the biggest.

SYSTRAN (System Translation), a private corporation financed by government contracts founded by Hungarian-born linguist and computer scientist Peter Toma, was the main supplier of automated translation technology and services in the United States.

In the 1950s, Toma became interested in machine translation while studying at the California Institute of Technology.

Around 1960, Toma moved to Georgetown University and started collaborating with other machine translation experts.

The Georgetown machine translation project, as well as SYSTRAN's initial contract with the United States Air Force in 1969, were both devoted to translating Russian into English.

That same year, at Wright-Patterson Air Force Base, the company's first machine translation programs were tested.

SYSTRAN software was used by the National Aeronautics and Space Administration (NASA) as a translation help during the Apollo-Soyuz Test Project in 1974 and 1975.

Shortly after, SYSTRAN was awarded a contract by the Commission of the European Communities to offer automated translation services, and the company has subsequently amalgamated with the European Commission (EC).

By the 1990s, the EC had seventeen different machine translation systems focused on different language pairs in use for internal communications.

In 1992, SYSTRAN began migrating its mainframe software to personal computers.

SYSTRAN Professional Premium for Windows was launched in 1995 by the company.

SYSTRAN continues to be the industry leader in machine translation.

METEO, which has been in use by the Canadian Meteorological Center in Montreal since 1977 for the purpose of translating weather bulletins from English to French; ALPS, developed by Brigham Young University for Bible translation; SPANAM, the Pan American Health Organization's Spanish-to-English automatic translation system; and METAL, developed at the University of Toronto.

In the late 1990s, machine translation became more readily accessible to the general public through web browsers.

Babel Fish, a web-based application created by a group of researchers at Digital Equipment Corporation using SYSTRAN machine translation technology, was one of the earliest online language translation services (DEC).

Thirty-six translation pairs between thirteen languages were supported by the technology.

Babel Fish began as an AltaVista web search engine tool before being sold to Yahoo! and then Microsoft.

The majority of online translation services still use rule-based and statistical machine translation.

Around 2016, SYSTRAN, Microsoft Translator, and Google Translate made the switch to neural machine translation.

103 languages are supported by Google Translate.

Predictive deep learning algorithms, artificial neural networks, or connectionist systems based after biological brains are used in neural machine translation.

Machine translation based on neural networks is achieved in two steps.

The translation engine models its interpretation in the first phase based on the context of each source word within the entire sentence.

The artificial neural network then translates the entire word model into the target language in the second phase.

Simply said, the engine predicts the probability of word sequences and combinations inside whole sentences, resulting in a fully integrated translation model.

The underlying algorithms use statistical models to learn language rules.

The Harvard SEAS natural language processing group, in collaboration with SYSTRAN, has launched OpenNMT, an open-source neural machine translation system.

Jai Krishna Ponnappan

You may also want to read more about Artificial Intelligence here.

See also: 

Cheng, Lili; Natural Language Processing and Speech Understanding.

Further Reading:

Arnold, Doug J., Lorna Balkan, R. Lee Humphreys, Seity Meijer, and Louisa Sadler. 1994. Machine Translation: An Introductory Guide. Manchester and Oxford: NCC Blackwell.

Bar-Hillel, Yehoshua. 1960. “The Present Status of Automatic Translation of Languages.” Advances in Computers 1: 91–163.

Garvin, Paul L. 1967. “Machine Translation: Fact or Fancy?” Datamation 13, no. 4: 29–31.

Hutchins, W. John, ed. 2000. Early Years in Machine Translation: Memoirs and Biographies of Pioneers. Philadelphia: John Benjamins.

Locke, William Nash, and Andrew Donald Booth, eds. 1955. Machine Translation of Languages. New York: Wiley.

Yngve, Victor H. 1964. “Implications of Mechanical Translation Research.” Proceedings of the American Philosophical Society 108 (August): 275–81.

Artificial Intelligence - What Is Deep Learning?


Deep learning is a subset of methods, tools, and techniques in artificial intelligence or machine learning.

Learning in this case involves the ability to derive meaningful information from various layers or representations of any given data set in order to complete tasks without human instruction.

Deep refers to the depth of a learning algorithm, which usually involves many layers.

Machine learning networks involving many layers are often considered to be deep, while those with only a few layers are considered shallow.

The recent rise of deep learning over the 2010s is largely due to computer hardware advances that permit the use of computationally expensive algorithms and allow storage of immense datasets.

Deep learning has produced exciting results in the fields of computer vision, natural language, and speech recognition.

Notable examples of its application can be found in personal assistants such as Apple’s Siri or Amazon Alexa and search, video, and product recommendations.

Deep learning has been used to beat human champions at popular games such as Go and Chess.

Artificial neural networks are the most common form of deep learning.

Neural networks extract information through multiple stacked layers commonly known as hidden layers.

These layers contain artificial neurons, which are connected independently via weights to neurons in other layers.

Neural networks often involve dense or fully connected layers, meaning that each neuron in any given layer will connect to every neuron of its preceding layer.

This allows the network to learn increasingly intricate details or be trained by the data passing through each subsequent layer.

Part of what separates deep learning from other forms of machine learning is its ability to work with unstructured data.

There are no pre-arranged labels or characteristics in unstructured data.

Deep learning algorithms can learn to link their own features with unstructured inputs using several stacked layers.

This is done by the hierarchical approach in which a deep multi-layered learning algorithm offers more detailed information with each successive layer, enabling it to break down a very complicated issue into a succession of lesser ones.

This enables the network to learn more complex information or to be taught by data provided via successive layers.

The following steps are used to train a network: Small batches of tagged data are sent over the network first.

The loss of the network is determined by comparing predictions to real labels.

Back propagation is used to compute and transmit any inconsistencies to the weights.

Weights are tweaked gradually in order to keep losses to a minimum throughout each round of predictions.

The method is repeated until the network achieves optimum loss reduction and high accuracy of accurate predictions.

Deep learning has an advantage over many machine learning approaches and shallow learning networks since it can self-optimize its layers.

Machine or shallow learning methods need human participation in the preparation of unstructured data for input, often known as feature engineering, since they only have a few layers at most.

This may be a lengthy procedure that takes much too much time to be profitable, particularly if the dataset is enormous.

As a result of these factors, machine learning algorithms may seem to be a thing of the past.

Deep learning algorithms, on the other hand, come at a price.

Finding their own characteristics requires a large quantity of data, which isn't always accessible.

Furthermore, as data volumes get larger, so do the processing power and training time requirements, since the network will be dealing with a lot more data.

Depending on the number and kinds of layers utilized, training time will also rise.

Fortunately, online computing, which lets anybody to rent powerful machines for a price, allows anyone to run some of the most demanding deep learning networks.

Convolutional neural networks need hidden layers that are not included in the standard neural network design.

Deep learning of this kind is most often connected with computer vision projects, and it is now the most extensively used approach in that sector.

In order to obtain information from an image, basic convnet networks would typically utilize three kinds of layers: convolutional layers, pooling layers, and dense layers.

Convolutional layers gather information from low-level features such as edges and curves by sliding a window, or convolutional kernel, over the picture.

Subsequent stacked convolutional layers will repeat this procedure over the freshly generated layers of low-level features, looking for increasingly higher-level characteristics until the picture is fully understood.

Different hyperparameters may be modified to find different sorts of features, such as the size of the kernel or the distance it glides over the picture.

Pooling layers enable a network to learn higher-level elements of an image in a progressive manner by down sampling the picture along the way.

The network may become too computationally costly without a pooling layer built amid convolutional layers as each successive layer examines more detailed data.

In addition, the pooling layer reduces the size of an image while preserving important details.

These characteristics become translation invariant, which means that a feature seen in one portion of an image may be identified in a totally other region of the same picture.

The ability of a convolutional neural network to retain positional information is critical for image classification.

The ability of deep learning to automatically parse through unstructured data to find local features that it deems important while retaining positional information about how these features interact with one another demonstrates the power of convolutional neural networks.

Recurrent neural networks excel at sequence-based tasks like sentence completion and stock price prediction.

The essential idea is that, unlike previous instances of networks in which neurons just transmit information forward, neurons in recurrent neural networks feed information forward while also periodically looping the output back to itself throughout a time step.

Recurrent neural networks may be regarded of as having a rudimentary type of memory since each time step includes recurrent information from all previous time steps.

This is often utilized in natural language processing projects because recurrent neural networks can handle text in a way that is more human-like.

Instead of seeing a phrase as a collection of isolated words, a recurrent neural network may begin to analyse the mood of the statement or even create the following sentence autonomously depending on what has already been stated.

In many respects akin to human talents, deep learning may give strong techniques of evaluating unstructured data.

Unlike humans, deep learning networks never get tired.

Deep learning may substantially outperform standard machine learning techniques when given enough training data and powerful computers, particularly given its autonomous feature engineering capabilities.

Image classification, voice recognition, and self-driving vehicles are just a few of the fields that have benefited tremendously from deep learning research over the previous decade.

Many new exciting deep learning applications will emerge if current enthusiasm and computer hardware upgrades continue to grow.

~ Jai Krishna Ponnappan

You may also want to read more about Artificial Intelligence here.

See also: 

Automatic Film Editing; Berger-Wolf, Tanya; Cheng, Lili; Clinical Decision Support Systems; Hassabis, Demis; Tambe, Milind.

Further Reading:

Chollet, François. 2018. Deep Learning with Python. Shelter Island, NY: Manning Publications.

Géron, Aurélien. 2019. Hands-On Machine Learning with Scikit-Learn, Keras and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. Second edition. Sebastopol, CA: O’Reilly Media.

Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. 2017. Deep Learning. Cambridge, MA: MIT Press.

Artificial Intelligence - What Is The Loebner Prize For Chatbots? Who Was Lili Cheng?

A chatbot is a computer software that communicates with people using artificial intelligence. Text or voice input may be used in the talks.

In certain circumstances, chatbots are also intended to take automatic activities in response to human input, such as running an application or sending an email.

Most chatbots try to mimic human conversational behavior, however no chatbot has succeeded in doing so flawlessly to far.


Chatbots may assist with a number of requirements in a variety of circumstances.

The capacity to save time and money for people by employing a computer program to gather or disseminate information rather than needing a person to execute these duties is perhaps the most evident.

For example, a corporation may develop a customer service chatbot that replies to client inquiries with information that the chatbot believes to be relevant based on user queries using artificial intelligence.

The chatbot removes the requirement for a human operator to conduct this sort of customer service in this fashion.

Chatbots may also be useful in other situations since they give a more convenient means of interacting with a computer or software application.

A digital assistant chatbot, such as Apple's Siri or Google Assistant, for example, enables people to utilize voice input to get information (such as the address of a requested place) or conduct activities (such as sending a text message) on smartphones.

In cases when alternative input methods are cumbersome or unavailable, the ability to communicate with phones by speech, rather than needing to type information on the devices' displays, is helpful.

Consistency is a third benefit of chatbots.

Because most chatbots react to inquiries using preprogrammed algorithms and data sets, they will often respond with the same replies to the same questions.

Human operators cannot always be relied to act in the same manner; one person's response to a query may differ from another's, or the same person's replies may change from day to day.

Chatbots may aid with consistency in experience and information for the users with whom they communicate in this way.

However, chatbots that employ neural networks or other self-learning techniques to answer to inquiries may "evolve" over time, with the consequence that a query given to a chatbot one day may get a different response from a question posed the next day.

However, just a handful chatbots have been built to learn on their own thus far.

Some, such as Microsoft Tay, have proved to be ineffective.

Chatbots may be created using a number of ways and can be built in practically any programming language.

However, to fuel their conversational skills and automated decision-making, most chatbots depend on a basic set of traits.

Natural language processing, or the capacity to transform human words into data that software can use to make judgments, is one example.

Writing code that can process natural language is a difficult endeavor that involves knowledge of computer science, linguistics, and significant programming.

It requires the capacity to comprehend text or speech from individuals who use a variety of vocabulary, sentence structures, and accents, and who may talk sarcastically or deceptively at times.

Because programmers had to design natural language processing software from scratch before establishing a chatbot, the problem of creating good natural language processing engines made chatbots difficult and time-consuming to produce in the past.

Natural language processing programming frameworks and cloud-based services are now widely available, considerably lowering this barrier.

Modern programmers may either employ a cloud-based service like Amazon Comprehend or Azure Language Understanding to add the capability necessary to read human language, or they can simply import a natural language processing library into their apps.

Most chatbots also need a database of information to answer to queries.

They analyze their own data sets to choose which information to provide or which action to take in response to the inquiry after using natural language processing to comprehend the meaning of input.

Most chatbots do this by matching phrases in queries to predefined tags in their internal databases, which is a very simple process.

More advanced chatbots, on the other hand, may be programmed to continuously adjust or increase their internal databases by evaluating how users have reacted to previous behavior.

For example, a chatbot may ask a user whether the answer it provided in response to a specific query was helpful, and if the user replies no, the chatbot would adjust its internal data to avoid repeating the response the next time a user asks a similar question.

Although chatbots may be useful in a variety of settings, they are not without flaws and the potential for abuse.

One obvious flaw is that no chatbot has yet been proven to be capable of perfectly simulating human behavior, and chatbots can only perform tasks that they have been programmed to do.

They don't have the same aptitude as humans to "think outside the box" or solve issues imaginatively.

In many cases, people engaging with a chatbot may be looking for answers to queries that the chatbot was not designed to answer.

Chatbots raise certain ethical issues for similar reasons.

Chatbot critics have claimed that it is immoral for a computer program to replicate human behavior without revealing to individuals with whom it communicates that it is not a real person.

Some have also stated that chatbots may contribute to an epidemic of loneliness by replacing real human conversations with chatbot conversations that are less intellectually and socially gratifying for human users.

Chatbots, on the other hand, such as Replika, were designed with the express purpose of providing lonely people with an entity to communicate to when real people are unavailable.

Another issue with chatbots is that, like other software programs, they might be utilized in ways that their authors did not anticipate.

Misuse could occur as a result of software security flaws that allow malicious parties to gain control of a chatbot; for example, an attacker seeking to harm a company's reputation might try to compromise its customer-support chatbot in order to provide false or unhelpful support services.

In other circumstances, simple design flaws or oversights may result in chatbots acting unpredictably.

When Microsoft debuted the Tay chatbot in 2016, it learnt this lesson.

The Tay chatbot was meant to teach itself new replies based on past discussions.

When users engaged Tay in racist conversations, Tay began making public racist or inflammatory remarks of its own, prompting Microsoft to shut down the app.

The word "chatbot" was first used in the 1990s as an abbreviated version of chatterbot, a phrase invented in 1994 by computer scientist Michael Mauldin to describe a chatbot called Julia that he constructed in the early 1990s.

Chatbot-like computer programs, on the other hand, have been around for a long time.

The first was ELIZA, a computer program created by Joseph Weizenbaum at MIT's Artificial Intelligence Lab between 1964 and 1966.

Although the software was confined to just a few themes, ELIZA employed early natural language processing methods to participate in text-based discussions with human users.

Stanford psychiatrist Kenneth Colby produced a comparable chatbot software called PARRY in 1972.

It wasn't until the 1990s, when natural language processing techniques had advanced, that chatbot development gained traction and programmers got closer to their goal of building chatbots that could participate in discussion on any subject.

A.L.I.C.E., a chat bot debuted in 1995, and Jabberwacky, a chatbot created in the early 1980s and made accessible to users on the web in 1997, both have this purpose in mind.

The second significant wave of chatbot invention occurred in the early 2010s, when increased smartphone usage fueled demand for digital assistant chatbots that could engage with people through voice interactions, beginning with Apple's Siri in 2011.

The Loebner Prize competition has served to measure the efficacy of chatbots in replicating human behavior throughout most of the history of chatbot development.

The Loebner Prize, which was established in 1990, is given to computer systems (including, but not limited to, chatbots) that judges believe demonstrate the most human-like behavior.

A.L.I.C.E, which won the award three times in the early 2000s, and Jabberwacky, which won twice in 2005 and 2006, are two notable chatbots that have been examined for the Loebner Prize.

Lili Cheng

Lili Cheng is the Microsoft AI and Research division's Corporate Vice President and Distinguished Engineer.

She is in charge of the company's artificial intelligence platform's developer tools and services, which include cognitive services, intelligent software assistants and chatbots, as well as data analytics and deep learning tools.

Cheng has emphasized that AI solutions must gain the confidence of a larger segment of the community and secure users' privacy.

Her group is focusing on artificial intelligence bots and software apps that have human-like dialogues and interactions, according to her.

The ubiquity of social software—technology that lets people connect more effectively with one another—and the interoperability of software assistants, or AIs that chat to one another or pass tasks to one another, are two further ambitions.

Real-time language translation is one example of such an application.

Cheng is also a proponent of technical education and training for individuals, especially women, in order to prepare them for future careers (Davis 2018).

Cheng emphasizes the need of humanizing AI.

Rather than adapting human interactions to computer interactions, technology must adapt to people's working cycles.

Language recognition and conversational AI, according to Cheng, are insufficient technical advancements.

Human emotional needs must be addressed by AI.

One goal of AI research, she says, is to understand "the rational and surprising ways individuals behave." Cheng graduated from Cornell University with a bachelor's degree in architecture."

She started her work as an architect/urban designer at Nihon Sekkei International in Tokyo.

She also worked in Los Angeles for the architectural firm Skidmore Owings & Merrill.

Cheng opted to pursue a profession in information technology while residing in California.

She thought of architectural design as a well-established industry with well-defined norms and needs.

Cheng returned to school and graduated from New York University with a master's degree in Interactive Telecommunications, Computer Programming, and Design.

Her first position in this field was at Apple Computer in Cupertino, California, where she worked as a user experience researcher and designer for QuickTime VR and QuickTime Conferencing in the Advanced Technology Group-Human Interface Group.

In 1995, she joined Microsoft's Virtual Worlds Group, where she worked on the Virtual Worlds Platform and Microsoft V-Chat.

Kodu Game Lab, an environment targeted at teaching youngsters programming, was one of Cheng's efforts.

In 2001, she founded the Social Computing group with the goal of developing social networking prototypes.

She then worked at Microsoft Research-FUSE Labs as the General Manager of Windows User Experience for Windows Vista, eventually ascending to the post of Distinguished Engineer and General Manager.

Cheng has spoken at Harvard and New York Universities and is considered one of the country's top female engineers 

~ Jai Krishna Ponnappan

You may also want to read more about Artificial Intelligence here.

See also: 

Cheng, Lili; ELIZA; Natural Language Processing and Speech Understanding; PARRY; Turing Test.

Further Reading

Abu Shawar, Bayan, and Eric Atwell. 2007. “Chatbots: Are They Really Useful?” LDV Forum 22, no.1: 29–49.

Abu Shawar, Bayan, and Eric Atwell. 2015. “ALICE Chatbot: Trials and Outputs.” Computación y Sistemas 19, no. 4: 625–32.

Deshpande, Aditya, Alisha Shahane, Darshana Gadre, Mrunmayi Deshpande, and Prachi M. Joshi. 2017. “A Survey of Various Chatbot Implementation Techniques.” Inter￾national Journal of Computer Engineering and Applications 11 (May): 1–7.

Shah, Huma, and Kevin Warwick. 2009. “Emotion in the Turing Test: A Downward Trend for Machines in Recent Loebner Prizes.” In Handbook of Research on Synthetic Emotions and Sociable Robotics: New Applications in Affective Computing and Artificial Intelligence, 325–49. Hershey, PA: IGI Global.

Zemčík, Tomáš. 2019. “A Brief History of Chatbots.” In Transactions on Computer Science and Engineering, 14–18. Lancaster: DEStech.

What Is Artificial General Intelligence?

Artificial General Intelligence (AGI) is defined as the software representation of generalized human cognitive capacities that enables the ...