Bar-Ilan University

About: Bar-Ilan University is a education organization based out in Ramat Gan, Israel. It is known for research contribution in the topics: Population & Poison control. The organization has 12835 authors who have published 34964 publications receiving 995648 citations. The organization is also known as: Bar Ilan University & BIU.

Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them

Abstract: Word embeddings are widely used in NLP for a vast range of tasks It was shown that word embeddings derived from text corpora reflect gender biases in society, causing serious concern Several recent works tackle this problem, and propose methods for significantly reducing this gender bias in word embeddings, demonstrating convincing results However, we argue that this removal is superficial While the bias is indeed substantially reduced according to the provided bias definition, the actual effect is mostly hiding the bias, not removing it The gender bias information is still reflected in the distances between “gender-neutralized” words in the debiased embeddings, and can be recovered from them We present a series of experiments to support this claim, for two debiasing methods We conclude that existing bias removal techniques are insufficient, and should not be trusted for providing gender-neutral modeling

Children's avoidance of lexical overlap: a pragmatic account.

Abstract: Children tend to choose an unfamiliar object rather than a familiar one when asked to find the referent of a novel name. This response has been taken as evidence for the operation of certain lexical constraints in children's inferences of word meanings. The present studies test an alternative—pragmatic—explanation of this phenomenon among 3-year-olds. In Study 1 children responded to a request for the referent of a novel label in the same way that they responded to a request for the referent of a novel fact. Study 2 intimated that children assume that labels are common knowledge among members of the same language community. Study 3 demonstrated that shared knowledge between a speaker and listener plays a decisive role in how children interpret a speaker's request. The findings suggest that 3-year-olds' avoidance of lexical overlap i s not unique t o naming and may derive from children's sensitivity t o speakers' communicative intentions. How do children learn the meaning of words so efficiently given the complexity o f the contexts i n which most new words are encountered? One prevailing answer to this question is that children's inferences about word meanings are guided by a set of internal lexical constraints, biases, or principles that allow children to bypass consideration of most of the logically plausible meanings of a word (Golinkoff, Mervis, & Hirsh-Pasek, 1994; Markman, 1989; Waxman, 1990). For instance, it is argued that children believe that words denote whole objects, that labels refer to categories of objects rather than to individual objects, and that every object has only one name. A second line of thinking points out that young children are knowledgeable about various communication practices and are sensitive to a number of pragmatic cues present in the discourse context that indicate a speaker's communicative intent (L. Bloom, 1998). For instance, young children attend to the direction of a speaker's eye gaze to establish the referent of the speaker's utterance (Baldwin, 1993), they are sensitive to a speaker's affective and behavioral expressions as indicative of whether his or her communicative intent was accomplished (Tomasello,

Stress and accessibility of proximity-related thoughts: exploring the normative and intraindividual components of attachment theory.

Abstract: Three studies examine the effects of stress on the accessibility of proximity-related thoughts. In all the studies, participants reported on their attachment style, and the accessibility of proximity themes and worries in a lexical decision task was assessed upon the priming of a stress or neutral word. In Study 2, the primed stress word was semantically related to attachment themes. In Study 3, lexical decisions were made under low or high cognitive load conditions. Overall, the priming of a stress word led to increased accessibility of proximity themes, regardless of attachment style. Anxious-ambivalent people also showed high accessibility to proximity themes and worries in both neutral and stress contexts. In most conditions, avoidant persons' reactions were similar to those of secure persons. However, they showed no accessibility to proximity worries even after the priming of a semantically related word and reacted with high accessibility to these worries upon the addition of cognitive load.

The affective component of the secure base schema: affective priming with representations of attachment security.

Abstract: Using an affective priming procedure (S. T. Murphy & R. B. Zajonc, 1993), 7 studies examined the effects of the contextual activation of representations of attachment security (secure base schema) on the evaluation of neutral stimuli under either neutral or stressful contexts. In all the studies, participants also reported on their attachment style. Results indicated that the subliminal priming of secure base representations led to more positive affective reactions to neutral stimuli than did the subliminal priming of neutral or no pictures under both neutral and stressful contexts. Although the subliminal priming of positively valued, attachment-unrelated representations heightened positive evaluations under neutral contexts, it failed to elicit positive affect under stressful contexts. The results also revealed interesting effects of attachment style. The discussion focuses on the affective component of the secure base schema.

Comparison of detrending methods for fluctuation analysis

Abstract: We examine several recently suggested methods for the detection of long-range correlations in data series based on similar ideas as the well-established Detrended Fluctuation Analysis (DFA). In particular, we present a detailed comparison between the regular DFA and two recently suggested methods: the Centered Moving Average (CMA) Method and a Modified Detrended Fluctuation Analysis (MDFA). We find that CMA performs the same as DFA in long data with weak trends and is slightly superior to DFA in short data with weak trends. When comparing standard DFA to MDFA we observe that DFA performs slightly better in almost all examples we studied. We also discuss how several types of trends affect different types of DFA. For weak trends in the data, the new methods are comparable with DFA in these respects. However, if the functional form of the trend in data is not a-priori known, DFA remains the method of choice. Only a comparison of DFA results, using different detrending polynomials, yields full recognition of the trends. A comparison with independent methods is recommended for proving long-range correlations.