Automatic evaluation of machine translation, based on computing n-gram similarity between system output and human reference translations, has revolutionized the development of MT systems. We explore the use of syntactic information, including constituent labels and head-modifier dependencies, in computing similarity between output and reference. Our results show that adding syntactic informatio...

Twitter conveys the opinions and interests of people in various topics and domains. In this paper we focus on political data with the purpose of automatically classifying opinion polarities and topic categories. Our experiments show that natural language processing techniques alone do not capture the information wanted. As a result, we use a hybrid method and we propose adding language features...

In this paper we extend our work described in (Dinu et al., 2011) by adding more conjugational rules to the labelling system introduced there, in an attempt to capture the entire dataset of Romanian verbs extracted from (Barbu, 2007), and we employ machine learning techniques to predict a verb’s correct label (which says what conjugational pattern it follows) when only the infinitive form is gi...

Factored gated restricted Boltzmann machine is a generative model, which capable to extract the transformation from an image pair. We extend this model by adding discriminative component, which allows directly use this model as a classifier, instead of using the hidden unit responses as features for another learning algorithm. To evaluate the capabilities of this model, we have created a synthe...

In this work we address the problem of manufacturing machine parts from sensed data. Constructing geometric models for objects from sensed data is the intermediate step in a reverse engineering manufacturing system. Sensors are usually inaccurate, providing uncertain sensed information. We construct geometric entities with uncertainty models from noisy measurements for the objects under conside...

This paper describes an approach to automatic extraction of verb meanings from machine-readable resources for the construction of large-scale knowledge sources. We describe semantic lters designed to reduce the number of incorrect assignments made by a purely syntactic technique. We report on our results of disambiguating the verbs in the semantic lters by adding WordNet sense annotations. 1 We...

The first and most popular is to work with meta-languages. For any language L we may form a new language by adding a predicate True, which acts only on sentences of L (i.e., sentences without the symbol True) and satisfies property 1. We can iterate this construction to obtain an infinite sequence of languages L1, L2, . . ., each of which contains the preceding language’s truth predicate. ∗UC B...

The goal of this study was to predict the genders of players of the online game Second Life using linguistic patterns from their chat data. This was accomplished using a rich set of stylistic features combined with various machine learning models. This project builds upon a previous study done at Stanford’s Virtual Human Interaction Lab in which very few linguistic features were used. Results s...

The SRI team joined the subtask of Chinese-English Patent machine translation evaluation, and submitted the transla­ tion results using a combined output from two types of gram­ mars supported in SRlnterp, with two different word seg­ mentations. We investigated the effect of adding sparse fea­ tures, together with several optimization strategies. Also,for the PatentMT domain, we carried out pr...

We present the result of a symbolic machine learning system, ALLiS 2.0 for the CoNLL-2001 shared task. ALLiS 2.0 is a theory reenement system using hierarchical data. Results are F=89.04 for subtask 1, F=68.02 for subtask 2 and F=67.70 for subtask 3 (development test). Adding manual rules improves considerably results specially for task 2 (F=79.44). For the test data, results are slightly worst...