When translating phrases (words or group of words), human translators, consciously or not, resort to different translation processes apart from the literal translation, such as Idiom Equivalence, Generalization, Particularization, Semantic Modulation, etc. Translators and linguists (such as Vinay and Darbelnet, Newmark, etc.) have proposed several typologies to characterize the different translation processes. However, to the best of our knowledge, there has not been effort to automatically classify these fine-grained translation processes. Recently, an English-French parallel corpus of TED Talks has been manually annotated with translation process categories, along with established annotation guidelines. Based on these annotated examples, we propose an automatic classification of translation processes at subsentential level. Experimental results show that we can distinguish non-literal translation from literal translation with an accuracy of 87.09%, and 55.20% for classifying among five non-literal translation processes. This work demonstrates that it is possible to automatically classify translation processes. Even with a small amount of annotated examples, our experiments show the directions that we can follow in future work. One of our long term objectives is leveraging this automatic classification to better control paraphrase extraction from bilingual parallel corpora.
—Distributed word representation techniques have been effectively integrated into the Information Retrieval retrieval task. The most basic approach to this is mapping a document and the query words into the vector space and calculating the semantic similarity between them. However, this has a bias problem towards documents with different lengths, which rank a small document higher compared to documents with larger vocabulary size. While averaging a document by mapping it into vector space, it allows each word to contribute equally, which results in increased distance between a query and the document vectors. In this paper, we propose that document length normalization should be applied to address the length bias problem while using embedding based ranking. Therefore, we have presented an experiment with traditional Length Normalization techniques over, word2vec (Skip-gram) model trained using the TREC Blog06 dataset for ad-hoc retrieval tasks. We have also attempted to include relevance signals introducing a simple Linear Ranking (LR) function, which considers the presence of query words in a document as evidence of relevancy while ranking. Our combined method of Length Normalization and LR significantly increases the Mean Average Precision up to 47% over a simple embeddings based baseline.
We present a method to detect topics in news articles. The topics of interest are each represented by a descriptive document. We train a model that can be seen as a similarity function between such a descriptive document and a news article. Our model is a neural network that operates on two kinds of inputs. (1) The full texts of the descriptive documents and the news articles are passed through the same recurrent encoder network and then the distance of the resulting encodings is taken. (2) Our proprietary NLP pipeline and knowledge base are used to recognize named entities and significant keywords and we compute features based on their overlap for a descriptive document and a news article. Our model finally combines the encoding distance with the overlap features and acts as a binary classifier. We evaluate and compare several model configurations on two datasets, a large one automatically created from Wikipedia and a smaller one created manually.
A Marching Cubes algorithm modification is presented in this work. This proposal is used to make from surface image discretized on volumetric pixels, called voxels, a three-dimensional representation from a flat image is raised. To achieve this effect, each component is connected continuously (without gaps) regardless of the values of the image surface. The aim is to achieve this effect, optimizing computational resources. The ambiguous elimination cases as well as the identification of non-trivial cases are among the main features attended. Once identified, the modelling process is optimized, seeking congruence with the edges of the volume, applying techniques that allow connecting the edges adjacent to the corners, generating the triangulation of the cube, and introducing a new set of edge equivalence classes and cube faces. The obtained results allow to observe the effective performance of the implemented modification.
The identication of indirect relationships between texts from diferent sources makes the task of text mining useful when the goal is to obtain the most valuable information from a set of texts. That is why in the eld of information retrieval the correct recognition of named entities plays an important role when extracting valuable information in large amounts of text. Therefore, it is important to propose techniques that improve the NER classiers in order to achieve the correct recognition of named entities. In this work, a graph structure for storage and enrichment of named entities is proposed. It makes use of synonyms and domain-specic ontologies in the area of computing. The performance of the proposed structure is measured and compared with other NER classiers in the experiments carried out.
In this chapter, we describe a lexical resource compiled by us and used in the experiments presented in Chapter 5.
