Sabine Ploux

A model for matching semantic maps between languages (French/English, English/French)

This article describes a spatial model for matching semantic values between two languages, French and English. Based on semantic similarity links, the model constructs a map that represents a word in the source language. Then the algorithm projects the map values onto a space in the target language. The new space abides by the semantic similarity links specific to the second language. Then the two maps are projected onto the same plane in order to detect overlapping values. For instructional purposes, the different steps are presented here using a few examples. The entire set of results is available at the following address: http://dico.isc.cnrs.fr.

Testing the cognitive relevance of a geometric model on a word association task: A comparison of humans, ACOM, and LSA

The general aim of this study is to validate the cognitive relevance of the geometric model used in the semantic atlases (SA). With this goal in mind, we compare the results obtained by the automatic contexonym organizing model (ACOM)—an SA-derived model for word sense representation based on contextual links—with human subjects’ responses on a word association task. We begin by positioning the geometric paradigm with respect to the hierarchical paradigm (WordNet) and the vector paradigm (latent semantic analysis [LSA] and the hyperspace analogue to language model). Then we compare ACOM’s responses with Hirsh and Tree’s (2001) word association norms based on the responses of two groups of subjects. The results showed that words associated by 50% or more of the Hirsh and Tree subjects were also proposed by ACOM (e.g., 71% of the words in the norms were also given by ACOM). Finally, we compare ACOM and LSA on the basis of the same association norms. The results indicate better performance for the geometric model.

Differentiating semantic categories during the acquisition of novel words: Correspondence analysis applied to event-related potentials

Growing evidence suggests that semantic knowledge is represented in distributed neural networks that include modality-specific structures. Here, we examined the processes underlying the acquisition of words from different semantic categories to determine whether the emergence of visual- and action-based categories could be tracked back to their acquisition. For this, we applied correspondence analysis (CA) to ERPs recorded at various moments during acquisition. CA is a multivariate statistical technique typically used to reveal distance relationships between words of a corpus. Applied to ERPs, it allows isolating factors that best explain variations in the data across time and electrodes. Participants were asked to learn new action and visual words by associating novel pseudowords with the execution of hand movements or the observation of visual images. Words were probed before and after training on two consecutive days. To capture processes that unfold during lexical access, CA was applied on the 100–400 msec post-word onset interval. CA isolated two factors that organized the data as a function of test sessions and word categories. Conventional ERP analyses further revealed a category-specific increase in the negativity of the ERPs to action and visual words at the frontal and occipital electrodes, respectively. The distinct neural processes underlying action and visual words can thus be tracked back to the acquisition of word-referent relationships and may have its origin in association learning. Given current evidence for the flexibility of language-induced sensory-motor activity, we argue that these associative links may serve functions beyond word understanding, that is, the elaboration of situation models.

Graph-Based Bilingual Word Embedding for Statistical Machine Translation

Bilingual word embedding has been shown to be helpful for Statistical Machine Translation (SMT). However, most existing methods suffer from two obvious drawbacks. First, they only focus on simple contexts such as an entire document or a fixed-sized sliding window to build word embedding and ignore latent useful information from the selected context. Second, the word sense but not the word should be the minimal semantic unit; however, most existing methods still use word representation. To overcome these drawbacks, this article presents a novel Graph-Based Bilingual Word Embedding (GBWE) method that projects bilingual word senses into a multidimensional semantic space. First, a bilingual word co-occurrence graph is constructed using the co-occurrence and pointwise mutual information between the words. Then, maximum complete subgraphs (cliques), which play the role of a minimal unit for bilingual sense representation, are dynamically extracted according to the contextual information. Consequently, correspondence analysis, principal component analyses, and neural networks are used to summarize the clique-word matrix into lower dimensions to build the embedding model. Without contextual information, the proposed GBWE can be applied to lexical translation. In addition, given contextual information, GBWE is able to give a dynamic solution for bilingual word representations, which can be applied to phrase translation and generation. Empirical results show that GBWE can enhance the performance of lexical translation, as well as Chinese/French-to-English and Chinese-to-Japanese phrase-based SMT tasks (IWSLT, NTCIR, NIST, and WAT).