Diana Trandabat

Grammar-Based Automatic Extraction of Definitions

The paper describes the development and usage of a grammar developed to extract definitions from documents. One of the most important practical usages of the developed grammar is the automatic extraction of definitions from web documents. Three evaluation scenarios were run, the results of these experiments being the main focus of the paper. One scenario uses an e-learning context and previously annotated e-learning documents; the second one involves a large collection of unannotated documents (from Wikipedia) and tries to find answers for definition type questions. The third scenario performs a similar question-answering task, but this time on the entire web using Google web search and the Google Translation Service. The results are convincing, further development as well as further integration of the definition extraction system in various related applications are already under way.

Minions at SemEval-2016 Task 4: or how to build a sentiment analyzer using off-the-shelf resources?

Minions, a team formed of first year students in the Master of Computational Linguistics, started the participation at Semeval-2016 as a semester project, aiming to build a model for analyzing and classifying “tweets” into positive, neutral and negative, according to the evoked sentiment, while getting familiar with Natural Language Processing tools and methods. Therefore, the backbone of our sentiment analyzer consists in several off-the-shelf, freely available resources, enhanced with a classifier trained on the SemEval-2016 data.

Mining Romanian texts for semantic knowledge

This papers presents a semantic role labeling system for Romanian texts. The semantic labeling system was developed using PASRL, a platform for supervised learning techniques. The developed platform tests several classifiers on different sub-problems of the SRL task (Predicate Identification, Predicate Sense Identification, Sense Identification, Argument Identification), chooses the ones with the greatest performance and returns a Semantic Role Labeling System (a sequence of trained models to run on new data).

Discourse theories vs. Topic-Focus articulation applied to prosodic focus assignment in Romanian

“…it is not the argument structure that triggers the intonational phrasing, but the [discourse, subclausal n.b.] relation of backgrounding.” (K. von Heusinger, 2007) Is it? If yes, how? The present paper, maintaining our attempts for applying Prague Schools Topic-Focus Articulation (TFA) algorithm on the syntax-prosody interface of Romanian, proposes two comparative lines of investigation for the intonational focus assignment: (A) The TFA algorithm is improved at clause level with hints from Gussenhovens SAAR (Sentence Accent Assignment Rule), then extended to inter-clause level, i.e. complex sentences. The new shape of the TFA algorithm is applied to compute the Topic-Focus values in the discursive context, while the information-structural (IS) spans of Theme(s)-Rheme(s) are detached, at clause level, as lowest-highest degrees of Communicative Dynamism (CD) vs. Systemic Order (SO). (B) The second approach we experiment for assigning intonational focus and phrasing is based on the combined and intensive use of discourse theories for computing the IS categories and structures: the Background-Kontrast entities (associated with the Prague Schools Topic-Focus) are obtained with Ashers (1993) Segmented Discourse Representation Theory (SDRT) analysis, while Theme-Rheme structures within the finite clause are computed with Leongs (2004) Inference-Boundary (IB) algorithm (of Hallidayan inspiration), applied for the first time to Romanian. Furthermore, this second direction is inspired from and joins the IS-discourse theory proposed by Heusinger (2007), which relies on SDRT inter-clausal evolution of discourse variables for computing the Background-Kontrast. While maintaining the classical SDRT (including rhetorical) discourse relations at the inter-clause level, Heusinger introduces a set of IS-semantics relations, and hands down at sub-clause level the rhetorical and focus particle relations with significant role in intonational-prosodic phrasing. Examples of these two types of research are compared to a gold, intonationally annotated set of Romanian sentences, the proposed theoretical and procedural techniques aiming to balance the pessimistic-realistic view on prosody prediction that it is the speaker-presupposition (and hearer-accommodation) which determines the IS focal scopes, rather than the bare text.

SentimentalITsts at SemEval-2016 Task 4: building a Twitter sentiment analyzer in your backyard.

The paper presents the system developed by the SentimentalITsts team for the participation in Semeval-2016 task 4, in the subtasks A, B and C. The developed system uses off the shelf solutions for the development of a quick sentiment analyzer for tweets. However, the lack of any syntactic or semantic information resulted in performances lower than those of

Integrating Contrastive Focus with Givenness and Topic-Comment: A hierarchical focus architecture of the Romanian Discourse-Prosody interface

This paper presents the following results: (a) On the basis of an extensive overview of the currently Information Structure (IS) theories, the first goal of our paper is to update the IS terminology for the three important IS dimensions: ± Givenness, Background-Focus (also referred as Topic-Focus or Backgroud-Kontrast), and Topic-Comment (also Theme-Rheme). (b) We propose an intonational discourse-level hierarchy among the Contrastive Focus (First Occurrence Focus), Second Occurrence Focus, Informational (Discourse-New) Focus, and Deaccented (Discourse-Given) Focus, while the phonetic properties of the considered intonational inequalities remain to be statistically established and weighted through speech analysis for Romanian. (c) This discourse-level prosodic hierarchy is combined, in a separate and independent way, with the clause-and phrase-level intonational hierarchies driven by Sentence Accent Assignment Rules, Nuclear Stress Rule, and the more recently Sentence Break Assignment Rules. (d) Based on the intonational focus hierarchies at points (b) and (c) above, a new architecture for the Discourse-Prosody interface is outlined, aiming to replace the classical approaches of Topic-Focus and Theme-Rheme algorithms (which can provide only incomplete Information Structure) for prosody prediction of Romanian. (e) The notions of explicit and implicit contrastive focus are defined, and the meaningful relevance of the contrastive intonation for the Romanian finite-clauses is pointed out by significant percentages of the contrastivity phenomena on George Orwells “1984” corpus. (f) Classes of examples illustrate and evaluate, for Romanian, the intonational-prosodic patterns of the contrastive and non-contrastive focus markers, categories, and domains.

The Curative Power of Medical Data

In an era when massive amounts of medical data became available, researchers working in biological, biomedical and clinical domains have increasingly started to require the help of language engineers to process large quantities of biomedical and molecular biology literature, patient data or health records. With such a huge amount of reports, evaluating their impact has long seized to be a trivial task. Linking the contents of these documents to each other, as well as to specialized ontologies, could enable access to and discovery of structured clinical information and foster a major leap in natural language processing and health research

Social media and the web of linked data

Written texts have perhaps never been so widely used as they are in todays social media context, with people constantly writing, sharing, commenting, getting involved. At the same time, Linked Data is emerging as an increasingly important topic, and research in this area has resulted in massive amounts of structured linguistic data. In this climate, we intend to analyze how linked data can help to structure and extract meaning from social medias short, informal and context dependent texts, with an emphasis on real-life applications.

SiadEnv Safety and Communication Features in Real Life Scenarios

SiadEnv system was designed to keep track of the energy consumes in residential and industrial buildings. This will analyze and compute the energy consume real need in various scenarios. The main objective of SiadEnv is to reduce the energy losses by taking action and modifying the room settings. Thus, SiadEnv computes the difference between outdoor and indoor temperature and adjusts the heating or cooling management in order to maintain the comfort index and to reduce the energy consume. Moreover, it contains indoor safety modules that prevent or reduce the impact of unwanted events such as flood, fire, motion control (thief entry etc). Due to SiadEnv modular design based on wireless sensors networks, the fire monitoring safety module can be easy reconfigured in order to extend its applications. As further work, the SiadEnv safety module will be redesigned into a new application with important social economic and environmental impact, which will use monitor forest fire and predict its dynamic, in order to provide crucial data for forest salvation.

Towards Extracting Semantic Information from Texts

This paper presents the general architecture of a system which creates a map of semantic information around a named entity (Person, Organization, etc.). Thus, after the user specifies a named entity, the system searches on the web and returns the first 200 web pages containing the specified entity, applies semantic roles on the returned paragraphs, and extracts a map of related actions involving the searched entity. This map of actions can then be chronologically ordered, thus illustrating the actions a certain entity has performed in a specific time frame (or at least the way it is reflected by the online world).