12-14 September
Hissar, Bulgaria
ranlp2011@lml.bas.bg
Ido Dagan (Bar Ilan University)
"Let computers think - in human language!"
Summary: A prominent aspect of thinking, and particularly of computational "thinking", is reasoning, or inference: the ability to derive new consequences from given information. Obviously, doing so requires a framework for meaning representations, over which inference would be applied. Traditionally, logic was introduced as the mathematical discipline for inference, and a plethora of logic languages were invented as meaning representations that support inference.
Indeed, following that vein, Natural Language Processing (NLP) traditionally prescribed translating natural language into logic representations first, and then conducting logical inferences over them. However, a seemingly more feasible alternative evolved in common practice NLP: applying inference directly over linguistic representations, which were originally invented to capture language structure rather than as explicit meaning representations (such as parse trees and lexical co-reference chains). In this talk I will examine some of these common-practice cases, and will elaborate on recent attempts to deliberately utilize common linguistic representations for broad inference. I will further suggest that we should invest in developing principled generic inference mechanisms over human-language representations. Such language-oriented mechanisms may end up being more effective for a broad range of inferences, while interfacing with "extra-linguistic" inferences where needed.
Patrick Hanks (University of the West of England, Bristol
and University of Wolverhampton)
"How people use words to make meanings"
Summary: This lecture will present a new theory of language, the Theory of Norms and Exploitations, which applies Prototype Theory to the large-scale empirical analysis of corpus data. One of the surprising findings of corpus linguistics during the past two decades has been that the regularities of patterns of linguistic behaviour are much more regular than generative linguists of the previous generation expected, while at the same time the deliberate irregularities (“exploitations of norms”) are much more irregular. An essential distinction must be made between possible usage and normal usage. A sane linguistic theory cannot hope to account for all possibilities of usage.
What is a linguistic pattern? Examples will be given showing that most patterns consists of a basic clause structure (verb with valencies) allied to collocational preferences. Valencies without collocations are insufficient to identify patterns accurately. Meanings are associated with patterns, not with words or sentences.
Patterns and meanings are probabilistic and statistically based, rather than deterministic.
The aim of this work is to provide a sound theoretical basis for creation of an unfashionably “knowledge-rich” infrastructure resource with a wide variety of potential applications in fields as different as computational linguistics and language teaching.
For further information, see my Corpus Pattern Analysis (CPA) project at http://nlp.fi.muni.cz/projects/cpa/.
Inderjeet Mani (Children's Organization of Southeast Asia)
"Getting Oriented: Spatial Prepositions, Frames of Reference, and Spatial Reasoning"
Summary: Humans are able to communicate effectively about the orientations of objects using spatial prepositions. The precise representation of the meaning of such prepositions has, however, been a subject of rich debate in both formal and cognitive linguistics. In this talk, I will examine the meaning of spatial prepositions from the standpoint of building artificial agents that reason about orientation based on qualitative spatial reasoning. From this point of view, objects can be viewed as geometric abstractions involving points, lines, and regions, and spatial prepositions involve specific relationships between these objects. Studies of speakers by Levinson (2003) and others across a wide variety of languages have revealed a basic inventory of coordinate systems (frames of reference) whose types are unevenly distributed across languages. I discuss the mapping of these frames of reference (intrinsic, absolute, and relative) to particular qualitative spatial reasoning frameworks, analyzing the semantics of particular classes of spatial prepositions in terms of the resulting grounded representations.
Roberto Navigli (Sapienza University of Rome)
" Is it just a waste of time? Word Sense Disambiguation for the skeptic."
Summary: Word Sense Disambiguation (WSD), the task of automatically associating meaning with words in context, is a long-standing problem in the field of computational linguistics. There can be no doubt the problem is a tough one. Researchers began to study the automatic association of meanings with words as long ago as the late 1940s. And they have been struggling to put their ideas into effective practice ever since. All too frequently their results have been disappointing not only in terms of disambiguation quality, but also when their WSD has been plugged into applications such as Information Retrieval and Machine Translation. Nevertheless, this pessimistic scenario has been progressively changing over the last decade, to the point that high disambiguation performance has been reported in recent work on the topic, indicating that WSD is more than alive. In this talk I will "challenge" the skeptic and analyze how and why WSD has achieved remarkable improvements in the last few years, and what promises it holds for the near future in terms of both in vitro performance and end-to-end applications.
Pierre-Paul Sondag (European Commission, DG INFSO)
" Language Technologies: A broad EU overview"
Summary: Language diversity constitutes an important cultural asset and has always been at the core of the European Union. The volume of digital content is growing to an amazing size and English plays less and less the role of lingua franca, more than 2/3 of Internet content, social media and electronic commerce transactions being now in other languages. One of the goals of the flagship initiative of the European Commission, /Digital Agenda for Europe/, is to make commercial and cultural services flow across Europe. The language technologies will be a key factor to addressing this challenge. The European Commission has a long track record in supporting language technologies since the 1970’s, it strongly reinforces that support since 2008 to enable language technology to benefit fully from developing science, cheaper computer power and language resources that are becoming more widely available now. A broad portfolio of projects is supported spanning from Machine Translation, speech processing, web-based services to language resource acquisition.
Hans Uszkoreit (University of Saarland)
" Research Results and Technology Visions for Multilingual Europe"
Summary: Despite measurable progress in multilingual and cross-lingual application research, our language technology will not be able soon to cope with the challenges posed by the multilingual setup of our continent. If this situation does not change, the preservation of linguistic diversity in the integrated European society could be in danger. Such a development would be sad since in this case Europe would not only sacrifice precious cultural heritage but also egalitarian and inclusive social ambition and highly promising commercial opportunity. In my talk I, will first summarize recent progress in European research on machine translation and other cross-lingual applications. The current state of the art will be assessed in the light of recognized social demands. I will then summarize the outcome of a collective brainstorming on technology visions for multilingual Europe. The European Network of Excellence META-NET has convened three expert panels in which representatives of research, development and business formulated powerful visions on future technologies. Each of these vision groups met three times. After sketching the outcome of this extensive intellectual deliberation, I will try to identify the results, directions and resources needed to realize the envisaged solutions. The keynote will end with a presentation of the META-NET process toward a strategic research agenda for European language technology. I will attempt to provoke the audience with a few personal observations and predictions but then also explain how every scientist, developer and research center can participate in the planning process.