Natural Language Processing (NLP) is a field of computer science and linguistics concerned with the interactions between computers and human (natural) languages; it began as a branch of artificial intelligence.

• S. Lakshmana Pandian

  What are morphologically rich languages?

  Word = root word + morpheme components (zero or many components)

  Recent replies ⋅ Show All (6)

  • 
    Gertrud Faaß replied

    Try looking at Southern African Bantu languages. Northern Sotho, for example, displays more than 300 possible inflection forms of verbs, and many ways to derivate (not even all known yet).Try looking at Southern African Bantu languages. Northern Sotho, for example, displays more than 300 possible inflection forms of verbs, and many ways to derivate (not even all known yet). Unfortunately, there are many ambiguous forms, too.

    4 days ago
• Kirubagari Balasubramaniam

  Neural networks for speech enhancement

  I'm working in the area of speech enhancement. I want to know how neural networks can be used to enhance the speech. After referring a paper I also tried to extract Mel scale cepstral coefficients.I'm working in the area of speech enhancement. I want to know how neural networks can be used to enhance the speech. After referring a paper I also tried to extract Mel scale cepstral coefficients. The function should be sinusoidal or tangential, so the output values should be '1' or '0'. How can I design a network? The input is Mel-frequency cepstral coefficients (MFCCs) of a frame of speech, what should be its output?

  Recent replies ⋅ Show All (5)

  • 
    Steven Longmire replied

    Depending on what you want to do with the end result, be it dictation or command and control for gauge entry in an oil field when you hands are full of grease or oil, Also, in very high noiseDepending on what you want to do with the end result, be it dictation or command and control for gauge entry in an oil field when you hands are full of grease or oil, Also, in very high noise environments in all, going back 2 decades, the masters of that were "Verbex Voice Systems", those were the days! Absolutely on the cepstral front end, and of course the traverse to fast Fourier did incredible things in a hybrid of Sphinx. I was very lucky to work with them for a few years, and still after so many years in almost 20 years actually, nothing yet has been able to match that success in extremely high noise environments. It all depends again on what you want to do do you want a phoneme-based recognizer, a sound pattern matching recognizer, or a combination of both? The reason I posed that question, is back in the day, I had the pleasure of choosing the route of going dictation (and dealing with the still prevalent problem of noise with phoneme-based recognizers), or command and control and constrained vocabulary for actual continuous speech for numerical data entry almost flawless at 99% or a little better in extremely high noise environments. The pleasure was a ride in a T 38 two-seat jet aircraft trainer for using speech recognition under high G loads and noise for avionics and weapons configuration control (everything of course except firing), the back of a refueling aircraft for verbal entry of international aircraft refueling, (fuel load, country the aircraft was from, and other pertinent information, all done in extremely high noise environments (military aircraft are not quiet! Like you find in civilian aircraft, the military ones are very noisy). Also fighting the extremely noisy M1 Abrams tanks with the hydraulics noise all I have to say on that is wow, it's no wonder so many went deaf because they did not wear ear protection, but the good thing is, the recognizers worked almost flawlessly in those environments. Lastly, on the pattern recognizer, the flight deck of an aircraft carrier, yes, it worked there extremely well. The interesting thing, and why I mention all of this is that outside of dictation like what you get with Dragon, or Microsoft's speech recognizers, the pattern matching algorithms were so strong (after almost 30 years of passion and speech technologies I keep abreast of what's out there), nothing has matched what Verbex had back then! To put an exclamation point behind that, three things come to mind that I was successful at being able to do, with one of them being funny as heck to bring some humor (that must be part of everything that I do these days), but I will say that till last. The first, had to deal with the recognition of sonar "audio" conversions for identification of friend or foe or maritime vehicle type. The reason this was done, is it takes a very long time to train a human how to recognize what it is they are hearing in the audio conversions of the sonar return signal. Again with the pattern recognizer and the algorithms that were developed by some incredible friends, it could differentiate was a whale, a boat, another submarine, or something else. There are other things in this day and age, but I'm telling you we had success at that over 20 years ago. The second one, was along the same lines, but had to do with Doppler 3-D radar ground-based signals coming back, again maybe after a few years of training a human could differentiate for example the difference between a Jeep or a truck, or even a human laying flat on the ground, and if you're ready for this, all on the back side of a hill, we could do that because it was limited and constrained what we were listening for and what we recognized. (You would think that that is something that would be used for defense everywhere, nope, just like Dragon has not had the adoption everyone thought it would have, or even Microsoft's speech recognizer, thanks to TV and science fiction movies people think when they buy recognition software that's what they're going to get. Unfortunately, as you all know, that does not exist (not yet anyway), and ironically enough living in Redmond Washington where I do, one of the people that through the years influenced me because I watched a Star Trek episode was (Scotty) when he picked up a mouse and gave the verbal command "Hello Computer". It was incredible that he happened to live in Redmond Washington as well, and he is sorely missed. The last one is funny as hell! I thought it was a joke when I was tasked to take care of the customer (once I found out it was real), and here it is. The customer was a farmer, and raised cows also for his dairy. The challenge was, there was different groups of cows that had been trained on different schedules of when to eat and when to make it into the dairy for milking. It took the farmer to control the gates for separating them as well as based on the time of day, which group of cows needed to go to the dairy, which was one direction with the gate, and which needed to go to the feed area. Yes, because cows make a certain sound when they're hungry, versus the sound they make when they go to the dairy for milking, it was actually very simple to capture those acoustic models, apply them to a controller for operating the opening and closing in the left or right format depending on where the cows needed to go with the recognizer. I don't know if he is still using it after all of these years but it was flawless, we were able to recognize cows :-) I appreciate some of the talent I have met through this online environment, and I do invite further discussion and drill down if anybody is interested in what they have read above, with a follow-on below on something that you may or may not know in the aspects of Verbex, and Dragon NaturallySpeaking. Which for the last 20+ years have been a passion and hobby that I got paid quite handsomely to utilize, develop, and especially to help people with disabilities. (Keep your eyes open in the coming year for something called "untapped resources", it is definitely related to speech recognition, and those unfortunate individuals without arms or hands. If anyone understands that statement, and wishes to volunteer to assist, I am most certainly looking forward to working with you, you will be joining a group of people that has no hidden agenda, no greed, and no politics. Just solely focused on helping people, and I can tell you folks, when you're able to help someone with what you do with speech recognition, the eye contact and the heartfelt "thank you" from them for absolutely changing their quality of life because you enable them to get work when no one else would even think about hiring them, is worth more than any monetary check you will ever receive in your life, and over the years I have received quite sizable ones. Here's the little history on mentioned earlier, there was a group back in the day my goodness over 30 years ago, that worked for Exxon oil company. They called the group "Verbal Exxon", when they split up that group (which by the way was formed to develop speech recognition for entering gauge readings instead of the manual way with hands in paper because their hands were full of oil all the time, they wanted to use speech recognition for recording gauge readings etc. When they closed that division down, the principal scientists split up and went to different directions, one, continuous speech (back then that was unheard of) but it was for command and control basic recognition, and numerical data entry. The company that was formed was Verbex. They still exist, but under a different name. The other group went the direction of dictation, of course I am speaking of Jim and Janet Baker, the original founders of Dragon NaturallySpeaking. I was only going to respond with a couple of sentences, but you know how it goes when thoughts start flooding in based on something that triggers them from your history, and most certainly this discussion triggered a lot, and there's a lot more I just simply do not want to bore you. The other players of course are Ray Kurzweil (if you have not read his books singularity and age of the spiritual machine, I highly recommend that you do because he was dead on target on a lot of things related to speech, artificial intelligence, and a whole lot more! I'm not trying to be a salesman on his books, it's just incredible reading that you should do if you are passionate about speech recognition and AI. I will tell you one thing that drives me absolutely nuts, is when someone reinvents the wheel, but really they didn't but they think they did, they just basically duplicated that which is been done before as some of the people have written prior to my response, it's not new! There are some things that are new, and you will see them appearing over this next year that are fascinating, it's being done utilizing open source technology so everyone can benefit and hopefully we stop this bull crap of everybody trying to make their gazillion dollars off of something that you know everybody will have in the future. Now there's nothing wrong with putting food in your refrigerator, but there's other ways to do it than putting everything into that stuff that's already been done before, just having a different wrapper on it. Stop the madness! Let's all come together make the next-generation happen, and as you know and I've even mentioned and what I have shared, pick a vertical, master it, and make your money that way, if you pick the right vertical market you will win, instead of trying to have the golden holy Grail speech recognizer that works for everything. Simply, that is still quite a few years off, but I can tell you with the engines and capabilities and things like GPU parallel processing (basically a supercomputer on your desktop, I personally like CUDA, Nvidia GPU's, I like the Tesla systems from them but they want too much money for them, to actually rock 'n roll with some neat stuff, if you can afford them get to watercooled GTX 690s that just came out in Quad SLI on a system that is only going to cost you around $2500 fully loaded, well I can't go too much further because I'm getting close to nondisclosure of a couple of things but, unless you have friends at rack space or friends in China where they're running 10,000 GPU's, you might be wise to get a couple of those video cards in Quad SLI, learn Cuda, or direct compute from Microsoft and start leveraging the power of a couple thousand cores doing work for you, while I think you get the message. I can be reached through Skype, my ID is: anti-terror or my direct phone: 425-329-3408 My best wishes to everyone for good health for you and your family, if you really want to do something, and money is not your motivator, but helping people is, I look forward to meeting you :-) All the best, Steven Longmire May 22, 2012

    5 days ago
• Michael Sun

  Are there any algorithms for automatic corpus annotation?

  I want to conduct a NER system. In terms of the huge corpus, I'm seeking an algorithm to automatically annotate the corpus. Clues and paper recommendation are also welcomed. Thank you all!

  Recent replies ⋅ Show All (2)

  • 
    Yogan Jaya Kumar replied

    Try look at General Architecture for Text Engineering (GATE) tool. It uses NLP framework that provides an easy to use platform to employ NER. Go through their manual to see how it works. You canTry look at General Architecture for Text Engineering (GATE) tool. It uses NLP framework that provides an easy to use platform to employ NER. Go through their manual to see how it works. You can also go to their homepage and refer to their publications.

    10 days ago
• Snigdha Paul

  Is there any sentence which does not follow CFG rule? Can any one give an example?

  Is it necessary to that all the sentences should follow CFG.

  Recent replies ⋅ Show All (16)

  • 
    Jacques Chauché replied

    La sémantique traverse la syntaxe : Je (mange, déguste, avale ...) cette pomme pourrie. Je (juge, pense, estime, ...) cette pomme pourrie. Si la nature du mot 'pourrie' est la même dans les deux casLa sémantique traverse la syntaxe : Je (mange, déguste, avale ...) cette pomme pourrie. Je (juge, pense, estime, ...) cette pomme pourrie. Si la nature du mot 'pourrie' est la même dans les deux cas sa position syntaxique diffère.

    14 days ago
• Farideh sh

  Could we have a Natural language to Conceptual language translator?

  We all know when we hear any statement in my Native language our brain convert this statement to Conceptual language that constructed in our brain. could we have artificial translator that do thisWe all know when we hear any statement in my Native language our brain convert this statement to Conceptual language that constructed in our brain. could we have artificial translator that do this conversion from any language, so whit this translator we can understand any Foreign language just like Native language?

  Recent replies ⋅ Show All (1)

  • 

    Comment protected

    Join ResearchGate now to read this comment.
• Nagendra Pratap Singh

  What are the research area of natural language processing

  Research area

  Recent replies ⋅ Show All (8)

  • 

    Comment protected

    Join ResearchGate now to read this comment.
• Lea Kosnik

  Does anyone know of any academic publications in legal literature that address NLP? Text analysis on business contracts, for example?

  My problem is that when you search on terms such as "textual analysis" in legal publication databases, you get a whole bunch of not relevant articles because "text analysis" has its own distinctMy problem is that when you search on terms such as "textual analysis" in legal publication databases, you get a whole bunch of not relevant articles because "text analysis" has its own distinct meaning in the legal field!

  Recent replies ⋅ Show All (2)

  • 
    Lea Kosnik replied

    That'd be great! Can you email them to me? kosnikl@umsl.edu

    27 days ago
• Kamil Czarnogórski

  What are the latest trends in short messages understanding (or analysis of user preferences based on short text)?

  What are the latest trends in short messages understanding (or analysis of user preferences based on short text)?
• AdityaPandu Wicaksoono

  What algorithm for natural language processing in Indonesian?

  I need some information about algorithm Text To Speech,..In this case the language is Indonesian,.Anybody give me some information, please,... Thank you,.

  Recent replies ⋅ Show All (4)

  • 
    Kalika Bali replied

    All the above are for Speech Recognition, not TTS. Please check Festival website http://www.cstr.ed.ac.uk/projects/festival/ for TTS. Are you looking to create a TTS for Bahasa Indonesia?

    Apr 17, 2012
• Waseem Alromimah

  We have a Corpus tagged for Arabic Language. How we can benefit from it to generate research ideas?

  The tagged Corpus we have for Arabic Language is for the Holy Quran.

  Recent replies ⋅ Show All (3)

  • 
    Johan Zahri replied

    cool, i kind a have a very vague road map.. the ultimate goal is to be able to interpret alif lam mim.. one of the milestones is to have a dictionary/encyclopedia/semantic web?/etymology directory ofcool, i kind a have a very vague road map.. the ultimate goal is to be able to interpret alif lam mim.. one of the milestones is to have a dictionary/encyclopedia/semantic web?/etymology directory of every terms in the quran.. don't know what the exact terminology to describe such a thing.. to get to that milestone maybe we need to make a conducive collaborative environment.. i think this environment must be able to have multi-lingual ability to translate between posts from one person in a language to another person in a different language.. the corpus and probably ontology with it could help in translating between the languages in aiding the forum discussions.. such as : in translating between the languages.. deriving closest match/suggestion to a term from one language to another based on sound and/or symbols construct, meaning/synonyms/antonyms, etc.. just spitting out ideas.. :D

    Apr 16, 2012
• Wu Liang

  A open source Conecpt Network (also called Semantic Network) browser named "ConceptBro" is released now. It maybe the first in the world to support all of "WordNet, WikiNet, JA-WordNet" . Since itA open source Conecpt Network (also called Semantic Network) browser named "ConceptBro" is released now. It maybe the first in the world to support all of "WordNet, WikiNet, JA-WordNet" . Since it has a abstract layer of database schema, it's very easy to extend to support other formats.

  

  Recent replies ⋅ Show All (1)

  • 
    Wu Liang replied

    It's website on github: http://wuliang.github.com/assets/projects/ConceptBro/

    

    Apr 14, 2012
• Ardini S Kartika

  Where can I find words defined by CICC tagset for Indonesian Language?

  I have a problem creating an Indonesian Language dictionary corpus based for NICETEXT system [Chapman 1997]. I have listed the words into a WORDLIST, but when move on to TWLIST, I can't pair theI have a problem creating an Indonesian Language dictionary corpus based for NICETEXT system [Chapman 1997]. I have listed the words into a WORDLIST, but when move on to TWLIST, I can't pair the words with their type based on CICC tagset words types because I can't find the references that show what words are include in each type.
• Ana Sofia Saldanha

  What is the impact of Slang and Taboo Language in the translation of literary texts? Is the vocabulary largely accepted? The editorial policies are updated according to this vocabulary?

  Recent replies ⋅ Show All (1)

  • 
    Ana Sofia Saldanha replied

    I posed this question as I have written my Masters Thesis on Slang and Taboo Language and its reception in the Portuguese context...however I found out that in Portugal most of these "words" are cutI posed this question as I have written my Masters Thesis on Slang and Taboo Language and its reception in the Portuguese context...however I found out that in Portugal most of these "words" are cut out from the texts or are softened...I would like to know some other opinions.

    Mar 28, 2012
• Daniel Loftus

  What is the biggest leap forward to have taken place in machine translation between languages in the last five years?

  Are there any innovations though that stand out from the steady progress being made with cloud computing and translator APIs (e.g., Google Translator and Microsoft Translator)?

  Recent replies ⋅ Show All (3)

  • 
    Alice Cheang replied

    I find something to agree with in both of the above posts. Translation engines are indeed developing rapidly and, with each jump in technology, they are also showing a certain amount of improvement.I find something to agree with in both of the above posts. Translation engines are indeed developing rapidly and, with each jump in technology, they are also showing a certain amount of improvement. But I fear that qualitative improvements (such as the ability to detect fine differences between words with similar/related meanings but not the same meaning) is not keeping pace with other kinds of improvement (such as speed of translation). With some languages, e.g., Chinese and Japanese, which do not use spaces to indicate the end of one word and the beginning of the next, there is additional room for error: depending on where the machine decides to put breaks in between words and phrases, incredible howlers can arise. In the end there is no way to replace the discriminating human translator-- yet.

    Mar 27, 2012
• Luke Dunn

  Can we use graph geometry to summarise text?

  I am working on text summarisation. I have had an interesting idea based around representing text as a concept map. Since the map is an artefact more easily manipulated by a computer, I thought itI am working on text summarisation. I have had an interesting idea based around representing text as a concept map. Since the map is an artefact more easily manipulated by a computer, I thought it sensible to look at the geometry or graph attributes of the structure, which a good graph library could handle (I am Python). It struck me that much can be seen in the concept map. Imagine extracting entities from text and representing relations between them as edges. Entities that were referred too quite often would be more connected. If you could make a threshold for minimum edge connectivity and strip away less connected (hence less relevant) entities then you might have a summary. This is identical, to explain it another way, to using a graph drawing algorithm to arrange the map in radial form with proximity-to-centre mapping to connectivity (hence relevance), and then cutting out the central subgraph along some arbitrary curve. it's a lossy compression process Sidetrack This might have a relationship to fractals too because a summary of the fractal is the primary self-similar shape, with no recursion depth, showing increasing recursion corresponds to greater detail in the text perhaps. There is a method for writing novels called the snowflake method where the book is built incrementally. Maybe this principle could be also reversed and used to ad-lib or unpack longer more explanatory text from a simple summary. because the comperssion was lossy this kind of system would need a knowledgebase too. apologies for such a vague treatment but its early days for me on this. does anyone know whether anyone else has ever used these ideas, and if so could anyone kindly point me in a further direction for development of the theme?

  Recent replies ⋅ Show All (7)

  • 
    Diego Molla Aliod replied

    You probably want to look at the textGraphs workshop for work on graph-based natural language processing: http://www.textgraphs.org/ In summarisation, look at work by Dragomir Radev:You probably want to look at the textGraphs workshop for work on graph-based natural language processing: http://www.textgraphs.org/ In summarisation, look at work by Dragomir Radev: http://www.summarization.com/ In summarization of medical text, including a graph output that incorporates relations between clinical concepts, look at SemRep: http://skr.nlm.nih.gov/papers/index.shtml#SemRep Good luck!

    Mar 25, 2012