User:Robert Wahlstedt/Proposed/Augmentative Communications
There are two classes of people with speech challenges, the first class consists of those who had trouble speaking since birth. They are aware of the fact they are more challenging to understand. One hypothesis on speaking challenges is the disorder is more prevalent in children because they are overwhelmed by the conceptualizing and the formatting tasks of language. Bell Labs engineer Robert Lucky estimated the cortex cannot take more than 50 bits per second. Mihaly Csikszentmihalyi estimated one apperception to take 1/15th of a second or 105 bits per second. Regardless of the figure a person has limited bandwidth or computing power. Because it takes a lot of concentration to come up with metaphors and categorize ideas with words, the quality of speech declines. Along with this speech has limited prosody referred to in this paper as words per minute, pitch variance, and stress of syllables. With the knowledge that they can be unintelligible, they compensate using repetition or choose words that does not contain the element that they have trouble with. They may do these substitutions or repetitions unconsciously. People with speech aphasia often have trouble with pronouns. A goal of Clear Audio is to find the use of pronouns and the repetition by using summarizing tools. These methods are discussed in section 2. Other people have difficulty speaking later in life due to a stroke may be confident that their pronunciation is the same as it was before their decline and may not incorporate either repetition or choice of words. The second case is discussed in section 3.
2.1 Observations from a cochlear implant In order for a contextual agent of software that recognizes speech, it must develop a cohesive hypothesis about what a speaker is saying. In the book, I Can Hear You Whisper by Lydia Denworth discusses the requirements of language for cochlear implant patients. This book suggests that processing sound is a multifaceted event including both contextual clues for perception and the stemming. Denworth introduces David Pisoni’s work on “visual speech”. Denworth discusses what is meant for a person to see things without cognition and gave this as a reason why her son has an implant starting at age three. While experimenting with Audacity, we realize how redundant speech is as opposed to music. A more sporadic sampling of voice is enough for voice while for music every sound we must accommodate for.
2.2 Summary generation for detecting repetitions A piece of literature is similar to a piece of music in that both contain themes that are repeated throughout the piece. While the BBC has a recording of Tchaikovsky’s 1812 overture that lasts 17 minutes, most people remember only the theme. In this paper, we discuss the methods how literature themes stand out from the rest of the piece through both exact repetition and supporting variations of the theme. Some scientists who study the brain call this grouping “chunking.” Many textbooks of text mining hold the author’s beliefs that word categories or parts of speech are important. These papers create summaries by deleting irrelevant words until only the nuclear core remains. However, these authors overlook context. A new approach seeks to use structural knowledge of pragmatics throigh latent semantic analysis. For example, to get to the best diagnosis of patients, it is important to take into account their entire medical histories and not just a few quotes. This method does analysis on the entire document before concluding so metaphors and words with multiple meanings are not confused. This paper proposes if using metadata, or data about the data, to distinguish repetition of the theme from supporting variations, also known as satellites. This is achievable by finding the roots of words through their origins, translating the word prefix, root, and suffix into metadata. The brain constantly encounters more information than it can digest. Oliver Sacks’ book Musicophilia discusses how Chinese speakers represent a higher than average percentage of people who have pure tone. Of these, the greatest were musicians who started learning their instrument before age six. This is because the tonal system of the Chinese language is a very important element and is necessary to distinguish words from each other. The brain of the Chinese speakers adapts so it is is accustomed to listening to other Chinese speakers. At the time of birth a person is born they can absorb much information from their senses including hearing. Around 80 percent of the energy they get from food goes to developing their brain. The book, We are our Brains, discusses one theory that a person uses the equivalent of $1500 for their entire life or 15 watts per hour. In order to come up with a computational model we need to ask the right questions. As they start building heuristics and finding out what is important and what is not then less energy goes to fuel for the brain. Dean Buonomano tells in his book, Brain Bugs: How the Brain’s Flaws Shape Our Lives, of how musicians dedicate larger than average part of their brain to muscular memory in their fingertips through preallocation. He goes on to say the brain does not consciously govern real estate in the nervous system so this was a subconscious process. Why do English speakers who are not musicians have such a seeming disadvantage? One dominant theory is that of the Hebbian Synapse that neurons that fire together wire together. In other words if two events are not linked by the human brain as related, the brain acts like a filter and events are never associated or learned with one another. Stanislas Dehaene takes this notion further and his book Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts discusses the history of the mind and body model. In the ancient world, philosophers like Descartes often thought that the heart was the headquarters of human thought. The soul or spirit was a detachable element from the physical body. This is similar to today’s notion between the difference between computer software and computer hardware. Later scientists developed interfaces for working with the injured brain to capture electrical signals. These in addition to fMRI have introduced new material under review as far as what a person can understand. Through the new methods, we can detect electrical signals of which the brain encodes information into neurons. We now have empirical evidence that people see other people through filters and this becomes imprinted on their brain based on prior experiences. For example, we know that when person has had a negative experience with big dogs, certain elements the perception system exaggerate such as teeth, when the person encountering future dogs. This imagry is related to visual speech.
2.3. Ambiguity of what is a Summary A recent survey done at the University of Illinois showed that if surveyors present a passage of text to a human subject they are likely to disagree on what passage is a summary. This is because culture makes the subject biased and the subjects project their preconceived notions on the passage. This shows the benefits of an artificial intelligence system that notices pronouns that a person who is overcome in emotions fails to recognize. In a study by James Pennebaker at the University of Texas Austin, hedescribes how natural language processors can detect who might be win an election. In his book, The Hidden Life of Pronouns he accounts of how aides told John Kerry to use the pronoun “I” less. Should a person be of the same political party as John Kerry, they might overlook the fact that he was using words in this format. Linguists call this connotation lexical semantics, the study of what words denotes.
2.4 The role of stemming in understanding speech as it relates to neuroscience Neuroscientists use the analogy of two cameras perceiving words. When a person reads a word like “unbuttoning” the human brain strips off the “un” and the “ing” and it is left with two morphemes, “but” and “ton.” There is a priming effect where a person is more inclined to think about a certain flow of words. For example, after casa, there sometimes is the word Blanca like the movie. This is why we understand differing homograph meaning. A grapheme is a letter or a series of letters that map a phoneme in the target language. Imagine a web of words that links all words in existence. In the book, The Neuroscience of Language introduces a word-related functional web. It is a synchronous firing chain or synfire chain. At first glance, automata might be similar to this web. However, a computer differs from a brain because even computers with multiple processing cores are asynchronous in the sense that the computer can only process a few threads unified by a central process. The human brain is a democracy having processes happening simultaneously and the strongest impulse, usually by a collective number of neurons fire that is what the human brain thinks. Miguel Nicolelis argues in his book, Beyond Boundaries, there is not a separate part of the brain language support. However, a certain sequence of neurons that light up when a person reads about an object because it invokes an emotion. The book, The First Word, describes how non-humans animals could use simple language. Instances of non-humans are territorial warning signs and primitive inner- species communication. The article “How Dolphins Say Hello” says that dolphins only whistle when they are in a group of dolphins and find themselves through echolocation. Grammar is a set of rules in a language that allows people to communicate and understand. There is a debate between B.F. Skinner and Noam Chomsky. Skinner believes that a person learns language through association, the sight of things along with the sound of the word. There conformity of this association matching constructs a dialect. Chomsky believes that a Martian scientist observing children in a single-language community would see that language is very similar regardless of the culture and therefore innate. Although reading is unnatural, our brain wants to see patterns and group objects together. We recognize lines and shapes that we can characters. We then group these characters into morphemes. We then put these morphemes into phrases, idioms, words, and sentences. These sentences form a paragraph. Similar to computers recognition is easy, recalling meaning is difficult.
The Language Wars by Henry Hitchings discusses how it is possible to use stemming given the evolution of human language. He says it is amazing that people spell words uniformly. Before computers, he points out that even dictionaries were inconsistent. There was a survey where a computer used phonetics to spell out words and 50 percent of the words did not agree with the phonetic spelling. Steven Pikner tells us that 84 percent of the words have spellings that conform patterns we can notice. As shown by these statistics, the ability to conform is possible. It is necessary to study of metadata to correlate etymology with words. Etymology is the study of finding word roots and morphology is the study of word making. Today advances in bio-informatics tools have yielded a complex study of the human genome and have made great strides in finding out mutations that can create a higher chance of cancer. We should do so because we can learn about the human race as well as expand our knowledge of words. Here are some examples: nickname originally came from nekename containing the compound eke and name. Eke came from ēac that means also. It means an extra name. The word “hobby” comes from the word hob and yn. Hob or hŏb means a threaded and fluted hardened steel cutter, resembling a tap, used in a lathe for forming the teeth of screw chasers, worm wheels (Webster’s Revised Unabridged Dictionary) and yn means “to be.” It means to spin time. The word “omelet” comes from the French word omelette. Omelette comes from alemette that comes from alemelle that as comes from la lemelle that means a thin plate like structure. We should know how to find the prefix, root, and suffix because we are able to understand why someone in history responded to situations. The written word often transcends a person's lifespan and the culture changes. A written manuscript is similar to a fossil. For example, rabbit comes from the words robète that means to steal. This explains why Mr. macGregor thinks Peter Rabbit as such a nuisance.
A computer's role in stemming Computers are able to carry out arithmetic or logical operations such as regular expressions or pattern matching. Given some rules and a lexicon file, they are able to study morphology.
“A” as in rag or Prague is big and “I” as in pin am small, “e” is somewhere in between. There has beesome big vowel shifting in England between 1350 and 1500 known as the big vowel change so words like big and small have the wrong letters. Little anlarge are right. Pimpf means "a little boy,” pimple means a little swelling, and pampers means a lot. Servant is in between.
It may seem everything the software needs are stemming or taking apart words to find their meaning. The book Found in Translation: How Language Shapes our Lives and Transforms the World, talks about how words are associated with different metaphors in different cultures. One example they give is intaxicato solar. This means food poisoning. It looks as though it means intoxicated as though someone had too much to drink. This is an example of what linguists call a “false friend" also known as a false cognate meaning that it looks as though they come from the same linguistic root but have different meaning. Another example is the word okoru in Japanese. Although it looks as though it may resemble the word to occur, it means to get angry. The book Found in Translation goes on to say that, certain brands like Mitsubishi have to rename their Pajero module and Honda Fitta module in particular countries. The countries have a different connotation of the words. The word gift means poison in German. The Online Etymology Dictionary says that it was often associated with prescribed medicine. People used the word gift as in a potion by a doctor and it came to mean something of tangible from a knowing person. Found in Translation tells how jokes in some languages are missed such as in the Harry Potter series Lord Voldemort is "Tom Marvolo Riddle" is an anagram for I am Lord Voldemort. The Bulgarian translation is Mersvoluko whose anagram translates to "And here I am, Lord Voldemort". The word in Hebrew originally was tohu meant formless and vohu meant empty. In French today toha bohu meant chaos and confusion. These words became associated with chaos later in Hebrew. Sometimes the environment makes it necessary to invent words to present the concepts behind them. For example to explain to the Hmong people about cancer at UC Davis, they compiled English to Hmong dictionary for medical terms. Martin Luther when translating the scriptures invented words so he can explain to people Latin concepts such as Machtwort (authoritative guidance). The book Eat, Shoots, and Leaves indicates this problem by having multiple definitions for the word shoots meaning to fire a fast projectile from a gun or shoots that are a part of a plant. The word leaves can also be from a plant or it can mean to abandon an area. These point out in addition to stemming the software needs to run on a distributed blackboard approach to artificial intelligence. A blackboard is comprehensive method that has several expert knowledge agents, they feed an aggregate stream with what the agent knows based on a limited perspective, and each agent fills in insufficiency of different agents. In our software the stemming is one agent and context of the previous spoken topic are another. In addition, features should consist of tone or pitch, tempo of words spoken or speed, and vibrancy or liveliness. The book Social Physics introduces a social-o meter that has a microphone for determining pitch and speaking duration. However, this microphone is unable to decide the real words. A key distinguishing reason of my research is that other software that can recognize speech such as Dragon Speaking Naturally Dictate has pride itself with how many sample voices have trained it. These speech recognition engines do not consider how important tonality is in making speech palatable. The brain needs some difference, dissonance, and variance to stay focused. When a robotic voice speaks, it is often takes greater amounts of concentration to listen. In today’s high-paced world, people might not always have energy enough to listen. Sounds constantly bombard with data. Those who hear a 60-hertz drone of a light fixture will likely ignore the drone because the brain has trained itself to ignore some sounds. Next, I hope to be able to find a feature set to annotate and make computer voices easier for people to listen.
In the Story of English in 100 Words, David Crystal discusses the origin of words. He gives the history of the word jail borrows from the French originating in the about the time of the Norman invasion of 1066 study we can tell because the word gaol is Latin. Crystalexplains that languages borrowed this word twice, being what Crystal refers to as double borrowed. We can tell this because the word did not get its origin meaning. This is also true for the word convey originality coming from the French word convoy. Our approach is to take this research and combine with what is known about UML, I* and the Zachman framework to discuss the questions of what, where, when, why, who and how. With this data, it is possible to verify that our summary correctly reflects the emphasis of the paper. In Empires of the World: A Language History of theWord, Nicholas Ostler demonstrates beginning with ancient Babylonians carrying off the Hebrew slaves and forcing them to melt into the culture of the Babylonians how language can melt together. John McWhorter writes this is what happened during the Norman invasion. There was not a bloody war instead slowly, over time the Normans introduced French. English, Crystal says, is like a vacuüm, sucking up words around it. McWhorter likens it to summer camp, where an exchange of ideas takes place. In Globish, Robert McCrum writes that non-native speakers to surpass the limitations of English by borrowing from other languages use globish. This field of anthropology is trans- cultural transfusion. This is an example of diffusion by choice. By studying hyperdiffusionism under the belief that man originated in one place, we can learn through cultural similarities. For instance, most civilized cultures today teach us that it is wrong to kill or commit adultery. Noam Chomsky writes about a universal grammar being indistinguishable by a Martian linguist. McWhorter claims that through DNA evidence anthropologists discovered the village where the Germanic dialect began. In tracing word etymologies, it is important to ask ourselves where features of a language began. Guy Deutscher makes the following observation: "Often, it is only the estrangement of foreign tongues, which their many exotic and outlandish features, that brings home the wonder of language's design." By incorporating origin into a NLP translator, we can see by parapraxis, images that speaker has in mind. Through a more careful of word choice, we have the advantage of seeing the speaker's words and come closer to making up for the disadvantage of not seeing him communicate those words to us. We can see how a language would transform and predict new words that might occur. By understanding how the introduction of new words and in what context it helps us predict what the likelihood that a speaker is, uttering these new words and this can greatly help the software that recognizes speech. 3. How to assist those with speech issues that do not overcompensate In people with speech issues who obtained them later in life due to a stroke it is important to incorporate elements of traditional speech recognition. These include language based game theory and eliminating what is not the topic. This paper proposes imitation of a biological system.
3.1 Language Based Game Theory A project in the Netherlands led by Jan Dietz seeks to map words as they relate to events in the world. Their premise is that we are like parrots repeating ideas that come to us through movies and books. In a game of broken telephone, a group of people sits in a circle and whisper a message to the person next to them. The message tends to vary as the first person conveys the message to the next person. These researchers claim through recursion we repeat these events demonstrated to us in a particular order. They believe they can forecast behaviors based on the past. The DEMO research group claims that BPMN, UML-AD, EPC, IDEF-3 are method independent. They take into account the redundant nature of business. It captures the conversations between actors. Jan Dietz explains that the subject chose to share one of their thoughts called the locutor. It appears as <locutor> : <illocution> : <addressee> : <fact> : <time> For example, let us consider checking a book out from a library. Certain circumstances must satisfy the checkout conditions. First, in order to check out books the subject should be a member at the library. Then he must not have any pending fees. Essence, the attributes that make the subject what he or she is. He is a candidate who can pick books out of the library. Should there be problems, he cannot do the transaction that is to pick up books. The production act of picking books out of the library is the executioner. It is a repeatable act. David Bellos tells an account of how his father was able to check into a hotel in a non-English speaking community because both he and the hotel staff were familiar with the process.
3.2 Eliminating what is not the Topic Another tool that can help with the speech is to find the topics not discussed. We can also use techniques learned in visual saliency to help us deduct clues. The book The Survival Game speaks of language like a stream of water. An adversary comes from the Latin word adversus meaning turned towards the flow. Language consists of hundreds or thousands of transactions and our program can learn what to expect. Language is much like the social behavior of ants. Can computers learn enough by watching transaction to find what stands out among the ordinary. The article Ant-Inspired Visual Saliency Detection, gives the image of ant behavior when they take the way that requires the least amount of effort to get towards their food. Saliency provides a method of processing of images for anything that stands out including edges. There is software for finding saliency.Consider the book, Where’s Waldo. In the Where’s Waldo example the stimulus is when we see something that resembles Waldo. For example, the viewer considers the color red with Waldo. Then on to the next level, they consider red stripes. Then they associate glasses. When a template is formed that confirmed by people of different cultures forms a grammar. It is possible by using a grammar to spot what the topic is not, much like a Bayesian spam filter. This monitors the flow of the conversation and when an unexpected sound appears in the recording it the software can quickly process the sound by assessing it and managing it.
3.3 Artificial Neural Networks As discussed before in this section automata is not an adequate algorithm for defining morpheme parsing. A neural network is a statistics algorithm used in machine learning. It is simple yet effective. Biological neural networks inspire artificial neural networks. They adaptable and self-organizing for real-time data. They also give fault tolerance. The brain forms the basis of an artificial neural network in a similar that an air plane can be compaired to a bird. They both have two wings and a tail and are both of the same air stream figures. When the inventors like Leonardo De Vinci was constructing a flying machine it very much resembled a bird. Similarly a neural network is a natural model of a brain (Abu-Mostafa). It is self-organizing, fault tolerant, and adaptable. We need to build a software design that resembles this working model. Looking at the PyBrain code it looks as though we can and should adopt something like this for our sentence processing sequence. In How the Mind Works, Steven Pinker discusses the benefits of using a neural network, however, linguists need to improve multiple instances of the same object.
