Research Institute for Artificial Intelligence

Romanian, Romania

Research Institute for Artificial Intelligence

Romanian, Romania
Time filter
Source Type

Denisleam Molomer S.,Polytechnic University of Bucharest | Trausan-Matu S.,Polytechnic University of Bucharest | Trausan-Matu S.,Research Institute for Artificial Intelligence
2016 20th International Conference on System Theory, Control and Computing, ICSTCC 2016 - Joint Conference of SINTES 20, SACCS 16, SIMSIS 20 - Proceedings | Year: 2016

This paper presents aspects of both oral and written communication disfluencies based on speech transcripts and chat conversations. The found results come from manual analyses of both audio transcriptions of students' verbalizations and XML codification of Computer Supported Collaborative Learning chat sessions. The aspects presented during the analyses are continuity, discontinuity, fluency and disfluency, these being selected with the purpose of being correlated. The entire analysis was based on emphasizing disfluency key words and concepts from dedicated taxonomies. © 2016 IEEE.

Tufis D.,Research Institute for Artificial Intelligence
Romanian Journal of Information Science and Technology | Year: 2016

Over the last twenty years or so, the approaches to part-of-speech tagging based on machine learning techniques have been developed or ported to provide high-accuracy morpho-lexical annotation for an increasing number of languages. Given the large number of morpho-lexical descriptors for a morphologically complex language, one has to consider ways to avoid the data sparseness threat in standard statistical tagging, yet to ensure that the full lexicon information is available for each wordform in the output. The paper overviews some of the major approaches to part-of-speech tagging and touches upon the tagset design, which is of crucial importance for the accuracy of the process.

Stuttgart/Berlin, 06-Dec-2016 — /EuropaWire/ — How is artificial intelligence (AI) influencing tomorrow’s mobility? How can we use its ideas today? How intelligent will the car of the future be? And above all: what role will people play in this? These were the central questions discussed by Mercedes-Benz experts at the fourth Future Talk in Berlin in dialogue with scientists, engineers and journalists. In the past few years the Mercedes-Benz Future Talk has dealt with the subjects of utopia, robotics and virtuality. This time, the focus was on the integration of artificial intelligence in the field of mobility and the interaction of humans and machines. Already in the 1960s researchers expected a major breakthrough in the development and application of artificial intelligence, but the human world still proved too complex for digital computers. However, due to the triumph of the internet, the mass of data that has become available with this and the huge increase in computing power of today’s computers, artificial intelligence is now entering people’s lives and also offers big opportunities and potential for the future of the automobile. “Artificial intelligence is a key future topic for Mercedes-Benz, in-car and beyond, such as in the fields of mobility services or in development and production”, says Anke Kleinschmit, Head of Daimler Group Research. “Artificial intelligence has ceased to be science fiction and the progress in autonomous driving is an impressive proof of this. Likewise, AI already assists the development phase and production by providing intuitive access to global knowledge and knowhow – Always tailored to the individual needs, experiences and knowledge of the employee. In addition to the technical development and data security, a basic prerequisite for the sustained success of artificial intelligence in all application cases is the acceptance by society and consumers. “Artificial intelligence will only be successful on a long-term basis if we succeed in building up trust between man and machine”, says futurologist Alexander Mankowsky. “We must define the division of tasks between human and artificial intelligence.” A necessary prerequisite is also to be aware of what artificial intelligence is able to do and what it isn’t. Because ultimately it always needs human participation and is based on human development. But it can support to make and examine decisions and therefore reach optimal results in a shorter space of time. The philosophy of Mercedes-Benz is – always to put human being at the centre of all activities. Cognitive vehicles: the car as a control centre for individualised AI An important objective of Mercedes-Benz’ activities relating to artificial intelligence is the development of cognitive vehicles. They are not only able to respond to certain situations; they even have enough knowledge about their environment to be able to act autonomously on this basis. Coupled with corresponding services they could become the fundament for a holistic mobility eco-system of the future. For example, they could autonomously analyse the current traffic situation for all forms of transport and draw up an individual mobility plan that suits the customer’s personal daily routine and mood. In addition, household robots and delivery drones could be linked to the system with the cognitive car as the control centre for this. Unlike smartphones and wearables, the car would surround the person and become a surrounding for a digital experience. It could analyse the driver’s behaviour, interpret needs and adapt accordingly. It would be able to identify what he or she wants in certain situations and what he or she needs. Examples of this are playing the right music to suit the current mood, setting the most pleasant temperature or developing services relating to health and safety. Moreover, the cognitive vehicle would offer self-determined access to an individualised artificial intelligence which supports human beings, entertains them and could even challenge them intellectually. Image and pattern recognition as an important milestone on the way to autonomous driving To successfully embark on this path, vehicles must be able to acquire knowledge about their environment as well as analyse it. This machine learning already plays an important role for autonomous driving as of today. Mercedes-Benz is working intensively on the further optimisation of automatic image and pattern recognition for driver assistance systems and autonomously driving vehicles. A decisive topic here is the interaction of cameras, sensors and the associated computing units. The system breaks down the pictures of road scenes into abstract segments with coloured marking. In this way it identifies buildings, vehicles, persons, trees and pavements among other things and reliably finds traffic lights as well as smaller dangerous obstructions on the road. Based on this, the autonomous vehicle analyses the traffic situation, predicts the behaviour of other road users and decides on its own behaviour. “In daylight many systems for image and pattern recognition, on the market are reliable”, says Dr Uwe Franke, responsible for image recognition/signal processing and sensor fusion in the Mercedes-Benz development department. “Meanwhile, our system even offers top level results at night and that is a major development. The next step is about recognising and interpreting people’s gestures and facial expressions.” It is the recognition of gestures, facial expressions and people’s understanding of machine behaviour that makes an operative interaction between man and autonomous vehicles possible at all. On this basis trust can be created between humans and machines. Vehicles must be able to make it clear that they recognise pedestrians and pay attention to them. Pedestrians must receive information about where an autonomous vehicle is going, how it will behave in the next few moments and how they should behave themselves. Finding and making faster use of ideas and potential with AI Mercedes-Benz is not just using artificial intelligence with regard to its vehicles. Among other things the company is testing self-learning systems in the observation of technology trends, in the interpretation of development and test data as well as for the industrial maintenance of its production and manufacturing facilities. Artificial intelligence can make a decisive contribution to diagnosing technical problems. For example, until now production maintenance staff either had to search through huge amounts of documents or fall back on their personal experience to get information about machine defects. The tested system handles documentation with natural language processing and serves as a semantic search engine. Unlike a keyword-based search engine the focus is on the meaning of the request. This enables requests in the form of various fault descriptions, for example “oil is leaking” or “leaking pipes”. In this way repair and maintenance processes can be speeded up and made more efficient. Mercedes-Benz to work with a leading AI institute in future Mercedes-Benz works with numerous renowned research institutions in the field of artificial intelligence. For years Daimler has been expanding its network to universities with innovative instruments such as “Forschungscampus”, tech centres, shared professorships and industry fellowships and start-up incubators and accelerators. In addition, the company will become a new member of the “MIT CSAIL Alliance Program” in the near future. With 50 research groups and around 1000 members of staff, the Computer Science and Artificial Intelligence Laboratory (CSAIL) of the Massachusetts Institute of Technology (MIT) is one of the leading institutes worldwide in the field of IT and AI. “Mercedes-Benz is rigorously advancing its research and development in different directions so as to continue to play a pioneering role in the automotive industry in the development and application of artificial intelligence. The new cooperation with the MIT ideally complements this. The partnership enables us to benefit even more directly from the research results of a leading world institute and to network with the best brains”, Anke Kleinschmit emphasises. “We aim to continue to play a leading role in shaping the future of mobility – with new mobility concepts, with cognitive cars and services which focus on people and make their daily lives easier and better.” As well as Anke Kleinschmit, Head of Daimler Group Research, the Daimler employees participating in this year’s round table discussion were futurologist Alexander Mankowsky, Dr Uwe Franke, responsible for image recognition/signal processing and sensor fusion and Patrick Klingler from IT Innovation Management. The external experts included Dr Miguel Nicolelis, Duke Center for Neuroengineering and Prof. Jürgen Schmidhuber, IDISIA Swiss Research Institute for Artificial Intelligence. Future Talk participant Prof. Jürgen Schmidhuber sees numerous fields of application for artificial intelligence in future. “It will take decades at most rather than centuries for us to develop true artificial intelligence. Artificial intelligences will learn almost everything that people can do – and much more besides. Their possibilities for shaping the future, are really only limited by our imagination.” Also Dr Miguel Nicolelis is convinced: “Artificial intelligence can lead to some sort of machine intelligence. This can be helpful in the future if it is under the control of human intelligence, not the other way around.” The Future Talk is a dialogue format successfully established by Mercedes-Benz in 2013. By exchanging ideas with vanguards from various disciplines, the brand shares its visions and, as the inventor of the automobile, demonstrates its expertise in shaping a desirable, mobile future. The focus topics to date are representative of the variety of this meta topic. This year’s Future Talk continues the discussion of the last few years on the subjects of “Utopias” (2013), “Robotics” (2014) and “Virtuality” (2015).

Mahloane M.J.,University of Witwatersrand | Trausan-Matu S.,Polytechnic University of Bucharest | Trausan-Matu S.,Research Institute for Artificial Intelligence
Proceedings - 2015 20th International Conference on Control Systems and Computer Science, CSCS 2015 | Year: 2015

The paper herein discusses the significance of metaphor annotation for a resource-scarce Bantu language of South Africa, Southern Sotho. In so doing, the need for development of NLP tools for this language and others like it will be indicated. Some of the challenges that have led to the lack of representation in NLP for this language include the absence of a corpus, and since this project refers to a recently compiled corpus for this language, it has been decided that the corpus needs to be annotated, in order to further prime language processing programs with its linguistic attributes. There are various reasons to annotate a corpus and various ways of annotating it, but this paper has chosen to focus on metaphor annotation. This sort of annotation can be performed at word or phrase/sentence level. For this project, the corpus will be annotated at word level. Another type of annotation that the corpus being worked with would require is word class tagging, which is the classification of word sense into their respective lexical classes. It may seem at first glance that metaphor annotation at word level is somehow the same as word tagging, however, metaphor annotation does not only identify a part-of-speech in its linguistic category, it includes semantic interpretation as well. This simultaneously disambiguates word senses. In addition, the computational requirements for such annotation will be looked into. There has already been challenges presented with word class tagging using conventional processing tools, and these indicated that this is due to the internal structure of Sesotho. However, metaphor annotation promises to open-up to other forms of language processing for Sesotho, which will in a long run make it untedious to process or for computer programs to recognize. © 2015 IEEE.

Ljubesic N.,University of Zagreb | Stefanescu D.,Research Institute for Artificial Intelligence | Tadic M.,University of Zagreb
Proceedings of the 10th Terminology and Knowledge Engineering Conference: New Frontiers in the Constructive Symbiosis of Terminology and Knowledge Engineering, TKE 2012 | Year: 2012

Although term extraction has been researched for more than 20 years, only a few studies focus on under-resourced languages. Moreover, bilingual term mapping from comparable corpora for these languages has attracted re-searchers only recently. This paper presents methods for term extraction, term tagging in documents, and bilingual term mapping from comparable corpora for four under-resourced languages: Croatian, Latvian, Lithuanian, and Romanian. Methods described in this paper are language independent as long as language specific parameter data is provided by the user and the user has access to a part of speech or a morpho-syntactic tagger.

Molomer S.,Polytechnic University of Bucharest | Trausan-Matu S.,Research Institute for Artificial Intelligence
Proceedings - RoEduNet IEEE International Conference | Year: 2014

This paper analyzes, from the polyphonic perspective used in the PolyCAFe system, the interactions among a group of high school students who, in parallel with a classroom style of interaction, use the instant messaging system Yahoo Messenger to develop a sorting algorithm. This type of analysis aims at finding the main topics in a chat conversation, the evaluation of utterances and their implicit links. Moreover, are investigated the ways in which the analysis can be automated using natural language processing tools. In addition to the previous research on using the polyphonic approach, the paper introduces the idea that pauses are important and should also be considered in the analysis. In collaborative hybrid (both face-to-face and online at a distance) learning educational environments one of the most used type of tools are the chat conversations. We can therefore consider Computer Supported Collaborative Learning (CSCL) as an important branch of hybrid learning. © 2014 IEEE.

Ciuca S.,Polytechnic University of Bucharest | Vlad A.,Polytechnic University of Bucharest | Vlad A.,Research Institute for Artificial Intelligence | Mitrea A.,Polytechnic University of Bucharest
UPB Scientific Bulletin, Series A: Applied Mathematics and Physics | Year: 2012

The paper focuses on a mathematical comparison between several single author corpora looking to give an answer to an open problem in literature: if and what are the terms one can speak of a general linguistic model or the author variability is too influent and we can only have separate author models. For the comparisons, an original procedure advanced by the authors in some previous studies was used, here extended and adapted for various forms of the corpora. That procedure implies the determination of the probability with a representative confidence interval for every investigated linguistic event in each analyzed corpus. The decision of determining the representative interval for probability is based on the probability estimation with statistical confidence intervals and also on tests verifying the hypothesis that the probability belongs to a certain interval. The final decision is also supported by the accuracy of the results considering the two types of error probability involved in the statistical tests. The experimental study is done on five independently built corpora, each of them being made of novels written by only one author. For each of them a detailed linguistic event analysis was made.

Boros T.,Research Institute for Artificial Intelligence | Dumitrescu S.D.,Research Institute for Artificial Intelligence
7th International ACM Conference on Management of Computational and CollEctive Intelligence in Digital EcoSystems, MEDES 2015 | Year: 2015

Currently, smartphones and tablets are firmly implanted within our daily lives. These devices have an entire ecosystem devoted to them, with applications and tools designed for their specifications: they use touch-enabled interfaces, have a limited amount of memory and CPU time available for apps (16/32MB limit on Android and iOS devices). A well-established research domain is the development of natural human-computer-interfaces (HCI) via voice and gestures. However, these interfaces are bound by the hardware resources available to them, and by the fact that they use network/Internet access to send/receive data, relying on dedicated servers for the decision making process. This paper focuses on the development of small robust deep-learning models that are designed to provide high quality text-to-speech (TTS) functionality (one of the three main components of HCI) on smart devices, without requiring network access. We obtain very good results in TTS text sub-tasks using models significantly smaller than those used in state-of-the-art approaches. © 2015 ACM.

Ion R.,Research Institute for Artificial Intelligence | Stefanescu D.,Research Institute for Artificial Intelligence
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2011

This paper presents an unsupervised word sense disambiguation (WSD) algorithm that makes use of lexical chains concept [6] to quantify the degree of semantic relatedness between two words. Essentially, the WSD algorithm will try to maximize this semantic measure over a graph of content words in a given sentence in order to perform the disambiguation. © 2011 Springer-Verlag.

Boros T.,Research Institute for Artificial Intelligence
International Conference Recent Advances in Natural Language Processing, RANLP | Year: 2013

General natural language processing and text-to-speech applications require certain (lexical level) processing steps in order to solve some frequent tasks such as lemmatization, syllabification, lexical stress prediction and phonetic transcription. These steps usually require knowledge of the word's lexical composition (derivative morphology, inflectional affixes, etc.). For known words all applications use lexicons, but there are always out-of-vocabulary (OOV) words that impede the performance of NLP and speech synthesis applications. In such cases, either rule based or data-driven techniques are used to automatically process these OOV words and generate the desired results. In this paper we describe how the above mentioned tasks can be achieved using a Perceptron with the Margin Infused Relaxed Algorithm (MIRA) and sequence labeling.

Loading Research Institute for Artificial Intelligence collaborators
Loading Research Institute for Artificial Intelligence collaborators