1,721,088 research outputs found
Dissecting Deep Language Models: The Explainability and Bias Perspective
Full text linkL'abstract è presente nell'allegato / the abstract is in the attachmen
Is it really that simple? Prompting Language Models for Automatic Text Simplification in Italian
Full text linkBuilding Bridges: A Dataset for Evaluating Gender-Fair Machine Translation into German
No full textThe translation of gender-neutral person-referring terms (e.g., the students) is often nontrivial. Translating from English into German poses an interesting case-in German, person-referring nouns are usually gender-specific, and if the gender of the referent(s) is unknown or diverse, the generic masculine (die Studenten (m.)) is commonly used. This solution, however, reduces the visibility of other genders, such as women and non-binary people. To counteract gender discrimination, a societal movement towards using gender-fair language exists (e.g., by adopting neosystems). However, gender-fair German is currently barely supported in machine translation (MT), requiring post-editing or manual translations. We address this research gap by studying gender-fair language in English-to-German MT. Concretely, we enrich a community-created gender-fair language dictionary and sample multi-sentence test instances from encyclopedic text and parliamentary speeches. Using these novel resources, we conduct the first benchmark study involving two commercial systems and six neural MT models for translating words in isolation and natural contexts across two domains. Our findings show that most systems produce mainly masculine forms and rarely gender-neutral variants, highlighting the need for future research. We release code and data at https://github.com/g8a9/building-b ridges-gender-fair-german-mt
PoliTeam @ AMI: Improving Sentence Embedding Similaritywith Misogyny Lexicons for Automatic Misogyny Identificationin Italian Tweets
Full text linken We present a multi-agent classification solution for identifying misogynous and aggressive content in Italian tweets. A first agent uses modern Sentence Embedding techniques to encode tweets and a SVM classifier to produce initial labels. A second agent, based on TF-IDF and Misogyny Italian lexicons, is jointly adopted to improve the first agent on uncertain predictions. We evaluate our approach in the Automatic Misogyny Identification Shared Task of the EVALITA 2020 campaign. Results show that TF-IDF and lexicons effectively improve the supervised agent trained on sentence embeddings.Presentiamo un classificatore multi-agente per identificare tweet italiani misogini e aggressivi. Un primo agente codifica i tweet con Sentence Embedding e una SVM per produrre le etichette iniziali. Un secondo agente, basato su TF-IDF e lessici misogini, è usato per coadiuvare il primo agente nelle predizioni incerte. Applichiamo la soluzione al task AMI della campagna EVALITA 2020. I risultati mostrano che TF-IDF e i lessici migliorano le performance del primo agente addestrato su sentence embedding
MilaNLP at SemEval-2022 Task 5: Using Perceiver IO for Detecting Misogynous Memes with Text and Image Modalities
Full text linkIn this paper, we describe the system proposed by the MilaNLP team for the Multimedia Automatic Misogyny Identification (MAMI) challenge. We use Perceiver IO as a multimodal late fusion over unimodal streams to address both sub-tasks A and B. We build unimodal embeddings using Vision Transformer (image) and RoBERTa (text transcript). We enrich the input representation using face and demographic recognition, image captioning, and detection of adult content and web entities. To the best of our knowledge, this work is the first to use Perceiver IO combining text and image modalities. The proposed approach outperforms unimodal and multimodal baselines
EFFECT OF MECHANICAL TRAUMA ON THE STAPEDIAL FOOTPLATE AFTER STAPEDOTOMY - A SCANNING ELECTRON-MICROSCOPIC STUDY
No full textThe possible occurrence of inner-ear trauma linked to stapedectomy was studied by scanning electron microscopy of the medial (or labyrinthine) side of human stapedial footplates after performing a hole with different instruments. The anatomic variations induced experimentally by the different procedures are presented in detail and discussed. Manual instruments were shown to induce irregularities on the rim of the hole, whilst the utilization of either electric or pneumatic drills produced more regular margins. Finally, laser-produced holes were also examined. Rather neat rims were observed, but the thermal effect produced by this device has to be considered the major parameter involved in a possible inner-ear postoperative trauma
3D cone beam(CBCT) in evaluation of frontal recess: Findings in youth population
No full textBACKGROUND, Frontal recess is the anatomical region most difficult to manage in endoscopic frontal sinus surgery due to the extreme variability of the cell patterns that may be observed in this area. CT has always been the gold standard in preoperative evaluation, but especially in the assessment of the causes of frontal recess obstruction and surgical failure. In recent years, this accredited and reliable method has been complemented by Computed Tomography Cone Beam (CBCT), which provides similarly detailed anatomical information with a lower dose of radiation. AIM, The purpose of this paper is to analyze and validate the use of CBCT in the study of frontal recess, and especially its anatomical variants in a youth population. MATHERIALS AND METHODS, We analyzed 500 CBCT images of paranasal sinuses of young subjects with sinus inflammation pathology between 2009 and 2011. RESULTS, We observed that the method is very sensitive in detecting anterior and posterior recess cells, also in a youth population and then report on some significant images. CONCLUSIONS, We confirm the validity of CBCT, which by virtue of its sensitivity and specificity may be used in the analysis of frontal recess pathologies, especially when a young population is involved
Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps
No full textCurrent automatic speech recognition (ASR) models are designed to be used across many languages and tasks without substantial changes. However, this broad language coverage hides performance gaps within languages, for example, across genders. Our study systematically evaluates the performance of two widely used multilingual ASR models on three datasets, encompassing 19 languages from eight language families and two speaking conditions. Our findings reveal clear gender disparities, with the advantaged group varying across languages and models. Surprisingly, those gaps are not explained by acoustic or lexical properties. However, probing internal model states reveals a correlation with gendered performance gap. That is, the easier it is to distinguish speaker gender in a language using probes, the more the gap reduces, favoring female speakers. Our results show that gender disparities persist even in state-of-the-art models. Our findings have implications for the improvement of multilingual ASR systems, underscoring the importance of accessibility to training data and nuanced evaluation to predict and mitigate gender gaps. We release all code and artifacts at https://github.com/g8a9/multilingual-asr-gender-gap
- …
