اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدColoring the Black Box: What Synesthesia Tells Us about Character Embeddings
98
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
In contrast to their word- or sentence-level counterparts, character embeddings are still poorly understood. We aim at closing this gap with an in-depth study of English character embeddings. For this, we use resources from research on grapheme-color synesthesia -- a neuropsychological phenomenon where letters are associated with colors, which give us insight into which characters are similar for synesthetes and how characters are organized in color space. Comparing 10 different character embeddings, we ask: How similar are character embeddings to a synesthetes perception of characters? And how similar are character embeddings extracted from different models? We find that LSTMs agree with humans more than transformers. Comparing across tasks, grapheme-to-phoneme conversion results in the most human-like character embeddings. Finally, ELMo embeddings differ from both humans and other models.
قيم البحث
اقرأ أيضاً
As the numbers of submissions to conferences grow quickly, the task of assessing the quality of academic papers automatically, convincingly, and with high accuracy attracts increasing attention. We argue that studying interpretable dimensions of thes
e submissions could lead to scalable solutions. We extract a collection of writing features, and construct a suite of prediction tasks to assess the usefulness of these features in predicting citation counts and the publication of AI-related papers. Depending on the venues, the writing features can predict the conference vs. workshop appearance with F1 scores up to 60-90, sometimes even outperforming the content-based tf-idf features and RoBERTa. We show that the features describe writing style more than content. To further understand the results, we estimate the causal impact of the most indicative features. Our analysis on writing features provides a perspective to assessing and refining the writing of academic articles at scale.
We explore how the expulsion of gas from star-cluster forming cloud-cores due to supernova explosions affects the shape of the initial cluster mass function, that is, the mass function of star clusters when effects of gas expulsion are over. We demon
strate that if the radii of cluster-forming gas cores are roughly constant over the core mass range, as supported by observations, then more massive cores undergo slower gas expulsion. Therefore, for a given star formation efficiency, more massive cores retain a larger fraction of stars after gas expulsion. The initial cluster mass function may thus differ from the core mass function substantially, with the final shape depending on the star formation efficiency. A mass-independent star formation efficiency of about 20 per cent turns a power-law core mass function into a bell-shaped initial cluster mass function, while mass-independent efficiencies of order 40 per cent preserve the shape of the core mass function.
Combining insights from both the effective field theory of quantum gravity and black hole thermodynamics, we derive two novel consistency relations to be satisfied by any quantum theory of gravity. First, we show that a particular combination of the
number of massless (light) fields in the theory must take integer values. Second, we show that, once the massless spectrum is fixed, the Wilson coefficient of the Kretschmann scalar in the low-energy effective theory is fully determined by the logarithm of a single natural number.
Using an artificial neutral network we explore the parameter space of supergravity grand unified models consistent with the combined Fermilab E989 and Brookhaven E821 data on $(g-2)_mu$. The analysis indicates that the region favored by the data is t
he one generated by gluino-driven radiative breaking of the electroweak symmetry. This region naturally leads to a split sparticle spectrum with light sleptons and weakinos but heavy squarks, with the stau and the chargino as the lightest charged particles. We show that if the entire deviation from the standard model $(g-2)_{mu}$ arises from supersymmetry, then supersymmetry is discoverable at HL-LHC and HE-LHC via production and decay of sleptons within the optimal integrated luminosity of HL-LHC and with a smaller integrated luminosity at HE-LHC.
When parsing morphologically-rich languages with neural models, it is beneficial to model input at the character level, and it has been claimed that this is because character-level models learn morphology. We test these claims by comparing character-
level models to an oracle with access to explicit morphological analysis on twelve languages with varying morphological typologies. Our results highlight many strengths of character-level models, but also show that they are poor at disambiguating some words, particularly in the face of case syncretism. We then demonstrate that explicitly modeling morphological case improves our best model, showing that character-level models can benefit from targeted forms of explicit morphological modeling.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
