György Szarvas

The BioScope corpus: biomedical texts annotated for uncertainty, negation and their scopes

BackgroundDetecting uncertain and negative assertions is essential in most BioMedical Text Mining tasks where, in general, the aim is to derive factual knowledge from textual data. This article reports on a corpus annotation project that has produced a freely available resource for research on handling negation and uncertainty in biomedical texts (we call this corpus the BioScope corpus).ResultsThe corpus consists of three parts, namely medical free texts, biological full papers and biological scientific abstracts. The dataset contains annotations at the token level for negative and speculative keywords and at the sentence level for their linguistic scope. The annotation process was carried out by two independent linguist annotators and a chief linguist – also responsible for setting up the annotation guidelines – who resolved cases where the annotators disagreed. The resulting corpus consists of more than 20.000 sentences that were considered for annotation and over 10% of them actually contain one (or more) linguistic annotation suggesting negation or uncertainty.ConclusionStatistics are reported on corpus size, ambiguity levels and the consistency of annotations. The corpus is accessible for academic purposes and is free of charge. Apart from the intended goal of serving as a common resource for the training, testing and comparing of biomedical Natural Language Processing systems, the corpus is also a good resource for the linguistic analysis of scientific and clinical texts.

What helps where – and why? Semantic relatedness for knowledge transfer

Remarkable performance has been reported to recognize single object classes. Scalability to large numbers of classes however remains an important challenge for todays recognition methods. Several authors have promoted knowledge transfer between classes as a key ingredient to address this challenge. However, in previous work the decision which knowledge to transfer has required either manual supervision or at least a few training examples limiting the scalability of these approaches. In this work we explicitly address the question of how to automatically decide which information to transfer between classes without the need of any human intervention. For this we tap into linguistic knowledge bases to provide the semantic link between sources (what) and targets (where) of knowledge transfer. We provide a rigorous experimental evaluation of different knowledge bases and state-of-the-art techniques from Natural Language Processing which goes far beyond the limited use of language in related work. We also give insights into the applicability (why) of different knowledge sources and similarity measures for knowledge transfer.

Automatic construction of rule-based ICD-9-CM coding systems

BackgroundIn this paper we focus on the problem of automatically constructing ICD-9-CM coding systems for radiology reports. ICD-9-CM codes are used for billing purposes by health institutes and are assigned to clinical records manually following clinical treatment. Since this labeling task requires expert knowledge in the field of medicine, the process itself is costly and is prone to errors as human annotators have to consider thousands of possible codes when assigning the right ICD-9-CM labels to a document. In this study we use the datasets made available for training and testing automated ICD-9-CM coding systems by the organisers of an International Challenge on Classifying Clinical Free Text Using Natural Language Processing in spring 2007. The challenge itself was dominated by entirely or partly rule-based systems that solve the coding task using a set of hand crafted expert rules. Since the feasibility of the construction of such systems for thousands of ICD codes is indeed questionable, we decided to examine the problem of automatically constructing similar rule sets that turned out to achieve a remarkable accuracy in the shared task challenge.ResultsOur results are very promising in the sense that we managed to achieve comparable results with purely hand-crafted ICD-9-CM classifiers. Our best model got a 90.26% F measure on the training dataset and an 88.93% F measure on the challenge test dataset, using the micro-averaged Fβ=1 measure, the official evaluation metric of the International Challenge on Classifying Clinical Free Text Using Natural Language Processing. This result would have placed second in the challenge, with a hand-crafted system achieving slightly better results.ConclusionsOur results demonstrate that hand-crafted systems – which proved to be successful in ICD-9-CM coding – can be reproduced by replacing several laborious steps in their construction with machine learning models. These hybrid systems preserve the favourable aspects of rule-based classifiers like good performance, and their development can be achieved rapidly and requires less human effort. Hence the construction of such hybrid systems can be feasible for a set of labels one magnitude bigger, and with more labeled data.

A multilingual named entity recognition system using boosting and c4.5 decision tree learning algorithms

In this paper we introduce a multilingual Named Entity Recognition (NER) system that uses statistical modeling techniques. The system identifies and classifies NEs in the Hungarian and English languages by applying AdaBoostM1 and the C4.5 decision tree learning algorithm. We focused on building as large a feature set as possible, and used a split and recombine technique to fully exploit its potentials. This methodology provided an opportunity to train several independent decision tree classifiers based on different subsets of features and combine their decisions in a majority voting scheme. The corpus made for the CoNLL 2003 conference and a segment of Szeged Corpus was used for training and validation purposes. Both of them consist entirely of newswire articles. Our system remains portable across languages without requiring any major modification and slightly outperforms the best system of CoNLL 2003, and achieved a 94.77% F measure for Hungarian. The real value of our approach lies in its different basis compared to other top performing models for English, which makes our system extremely successful when used in combination with CoNLL modells.

Cross-genre and cross-domain detection of semantic uncertainty

Uncertainty is an important linguistic phenomenon that is relevant in various Natural Language Processing applications, in diverse genres from medical to community generated, newswire or scientific discourse, and domains from science to humanities. The semantic uncertainty of a proposition can be identified in most cases by using a finite dictionary (i.e., lexical cues) and the key steps of uncertainty detection in an application include the steps of locating the (genre- and domain-specific) lexical cues, disambiguating them, and linking them with the units of interest for the particular application (e.g., identified events in information extraction). In this study, we focus on the genre and domain differences of the context-dependent semantic uncertainty cue recognition task.We introduce a unified subcategorization of semantic uncertainty as different domain applications can apply different uncertainty categories. Based on this categorization, we normalized the annotation of three corpora and present results with a state-of-the-art uncertainty cue recognition model for four fine-grained categories of semantic uncertainty.Our results reveal the domain and genre dependence of the problem; nevertheless, we also show that even a distant source domain data set can contribute to the recognition and disambiguation of uncertainty cues, efficiently reducing the annotation costs needed to cover a new domain. Thus, the unified subcategorization and domain adaptation for training the models offer an efficient solution for cross-domain and cross-genre semantic uncertainty recognition.

Combining query translation techniques to improve cross-language information retrieval

In this paper we address the combination of query translation approaches for cross-language information retrieval (CLIR). We translate queries with Google Translate and extend them with new translations obtained by mapping noun phrases in the query to concepts in the target language using Wikipedia. For two CLIR collections, we show that the proposed model provides meaningful translations that improve the strong baseline CLIR model based on a top performing SMT system.

Tune and mix: learning to rank using ensembles of calibrated multi-class classifiers

In subset ranking, the goal is to learn a ranking function that approximates a gold standard partial ordering of a set of objects (in our case, a set of documents retrieved for the same query). The partial ordering is given by relevance labels representing the relevance of documents with respect to the query on an absolute scale. Our approach consists of three simple steps. First, we train standard multi-class classifiers (AdaBoost.MH and multi-class SVM) to discriminate between the relevance labels. Second, the posteriors of multi-class classifiers are calibrated using probabilistic and regression losses in order to estimate the Bayes-scoring function which optimizes the Normalized Discounted Cumulative Gain (NDCG). In the third step, instead of selecting the best multi-class hyperparameters and the best calibration, we mix all the learned models in a simple ensemble scheme.Our extensive experimental study is itself a substantial contribution. We compare most of the existing learning-to-rank techniques on all of the available large-scale benchmark data sets using a standardized implementation of the NDCG score. We show that our approach is competitive with conceptually more complex listwise and pairwise methods, and clearly outperforms them as the data size grows. As a technical contribution, we clarify some of the confusing results related to the ambiguities of the evaluation tools, and propose guidelines for future studies.

Combining language sources and robust semantic relatedness for attribute-based knowledge transfer

Knowledge transfer between object classes has been identified as an important tool for scalable recognition. However, determining which knowledge to transfer where remains a key challenge. While most approaches employ varying levels of human supervision, we follow the idea of mining linguistic knowledge bases to automatically infer transferable knowledge. In contrast to previous work, we explicitly aim to design robust semantic relatedness measures and to combine different language sources for attribute-based knowledge transfer. On the challenging Animals with Attributes (AwA) data set, we report largely improved attribute-based zero-shot object class recognition performance that matches the performance of human supervision.

Linguistic scope-based and biological event-based speculation and negation annotations in the BioScope and Genia Event corpora

BackgroundThe treatment of negation and hedging in natural language processing has received much interest recently, especially in the biomedical domain. However, open access corpora annotated for negation and/or speculation are hardly available for training and testing applications, and even if they are, they sometimes follow different design principles. In this paper, the annotation principles of the two largest corpora containing annotation for negation and speculation – BioScope and Genia Event – are compared. BioScope marks linguistic cues and their scopes for negation and hedging while in Genia biological events are marked for uncertainty and/or negation.ResultsDifferences among the annotations of the two corpora are thematically categorized and the frequency of each category is estimated. We found that the largest amount of differences is due to the issue that scopes – which cover text spans – deal with the key events and each argument (including events within events) of these events is under the scope as well. In contrast, Genia deals with the modality of events within events independently.ConclusionsThe analysis of multiple layers of annotation (linguistic scopes and biological events) showed that the detection of negation/hedge keywords and their scopes can contribute to determining the modality of key events (denoted by the main predicate). On the other hand, for the detection of the negation and speculation status of events within events, additional syntax-based rules investigating the dependency path between the modality cue and the event cue have to be employed.

Prior art search using international patent classification codes and all-claims-queries

In this paper, we describe the system we developed for the Intellectual Property track of the 2009 Cross-Language Evaluation Forum. The track addressed prior art search for patent applications. We used the Lucene library to conduct experiments with the traditional TFIDF-based ranking approach, indexing both the textual content and the IPC codes assigned to each document. We formulated our queries by using the title and claims of a patent application in order to measure the (weighted) lexical overlap between topics and prior art candidates. We also formulated a language-independent query using the IPC codes of a document to improve the coverage and to obtain a more accurate ranking of candidates. Using a simple model, our system remained efficient and had a reasonably good performance score: it achieved the 6th best Mean Average Precision score out of 14 participating systems on 500 topics, and the 4th best score out of 9 participants on 10,000 topics.