Large, pre-trained transformer language models, which are pervasive in natural language processing tasks, are notoriously expensive to train. To reduce the cost of training such large models, prior work has developed smaller, more compact models whic
h achieves a significant speedup in training time while maintaining competitive accuracy to the original model on downstream tasks. Though these smaller pre-trained models have been widely adopted by the community, it is not known how well are they calibrated compared to their larger counterparts. In this paper, focusing on a wide range of tasks, we thoroughly investigate the calibration properties of pre-trained transformers, as a function of their size. We demonstrate that when evaluated in-domain, smaller models are able to achieve competitive, and often better, calibration compared to larger models, while achieving significant speedup in training time. Post-hoc calibration techniques further reduce calibration error for all models in-domain. However, when evaluated out-of-domain, larger models tend to be better calibrated, and label-smoothing instead is an effective strategy to calibrate models in this setting.
We present an algorithm based on multi-layer transformers for identifying Adverse Drug Reactions (ADR) in social media data. Our model relies on the properties of the problem and the characteristics of contextual word embeddings to extract two views
from documents. Then a classifier is trained on each view to label a set of unlabeled documents to be used as an initializer for a new classifier in the other view. Finally, the initialized classifier in each view is further trained using the initial training examples. We evaluated our model in the largest publicly available ADR dataset. The experiments testify that our model significantly outperforms the transformer-based models pretrained on domain-specific data.
We consider the problem of using observational data to estimate the causal effects of linguistic properties. For example, does writing a complaint politely lead to a faster response time? How much will a positive product review increase sales? This p
aper addresses two technical challenges related to the problem before developing a practical method. First, we formalize the causal quantity of interest as the effect of a writer's intent, and establish the assumptions necessary to identify this from observational data. Second, in practice, we only have access to noisy proxies for the linguistic properties of interest---e.g., predictions from classifiers and lexicons. We propose an estimator for this setting and prove that its bias is bounded when we perform an adjustment for the text. Based on these results, we introduce TextCause, an algorithm for estimating causal effects of linguistic properties. The method leverages (1) distant supervision to improve the quality of noisy proxies, and (2) a pre-trained language model (BERT) to adjust for the text. We show that the proposed method outperforms related approaches when estimating the effect of Amazon review sentiment on semi-simulated sales figures. Finally, we present an applied case study investigating the effects of complaint politeness on bureaucratic response times.
In deployment, systems that use speech as input must make use of automated transcriptions. Yet, typically when these systems are evaluated, gold transcriptions are assumed. We explicitly examine the impact of transcription errors on the downstream pe
rformance of a multi-modal system on three related tasks from three datasets: emotion, sarcasm, and personality detection. We include three separate transcription tools and show that while all automated transcriptions propagate errors that substantially impact downstream performance, the open-source tools fair worse than the paid tool, though not always straightforwardly, and word error rates do not correlate well with downstream performance. We further find that the inclusion of audio features partially mitigates transcription errors, but that a naive usage of a multi-task setup does not.
Laboratory experiments were conducted to study the effect of the aqueou extract of the
vegetative parts of the Euphorbia paralias at different concentrations (2%,4%,8%) in seed
germination and seeding growth of the Lepidium sativum, Lactuca sativa
and
Portulaca oleracea. The results indicated increase in germination ratio of Portulaca
oleracea. up to 7%. while the effect of the extract was not evident in the germination of the
seeds of Lepidium sativum, and decreased the germination ratio of Lactuca sativa
seeds by 35.9% at the concentration of 8%.In contrast, the extracts stimulated the length of
the stalks of Portulaca oleracea , Lepidium sativum and Lactuca sativa plants at all
used concentrations used. The highest rate of stimulated stalks legnth of Lepidium
sativum at 2% was 58.2%. inhibitory effect of water extract observed for root length of
Portulaca oleracea and Lactuca sativa plants at 2% , and stimulation effect pour
Lepidium sativum, The effect became evident at the concentrations of 4% and 8% for the
three plants.
The most sensitive and affected plants was Portulaca oleracea, where the length of the
root decreased by 53.38% at the concentration of 4% and by 72% at the concentration of
8% ,while the length of the root Lepidium sativum and Lactuca sativa decreased by
25.82% and 34.95% was decreased for 4% and 8%respectively.
The results suggested that this weed may affect seedling growth, due to inhibitory or
stimulatory effect of Allelochemicals , which present in water extract of this weed . and
may used as potential Bioherbicide after further experiments.
This study comes to reveal about the psychological and social effects of the use of
special young university for social networking sites, and by knowing the motives of
undergraduates to use these sites, and the various factors that led them to enga
ge in, as well
as to identify the psychological and social effects of the use of specifically Facebook site,
through a study field on a sample of university students Tishreen users of Facebook, where
we relied on the analytical descriptive approach and we used the questionnaire to collect
data from a sample of the tool (150) students from the University of Tishreen.
The research aims to expose the of The Educational Effects Social
Connections Network on University Youth. This research depended
on the descriptive approach. The questionnaire was toll used to
collect data. The questionnaire content (26) terms in three axis, they
are about effects: (personality, social and cultural).
This work theoretically investigates the nonlinear behaviour of
reinforced concrete dee aimed topcantilever beams with concentrated loads
at their free ends The study is aimed to in investigate the behaviour
and respnse of such deep cantilever bea
ms, and to help structural
engineers to design and adopt appropriate reinforcement detailing of
such elements. A complete review of literature on this subject is
made.
In this research, the danger and the horizontal force effects on the
engineering structures will be explained, and we will focus on the
horizontal winds force effects on the bridges, the high buildings and
special engineering structures. The resul
ted loads will be analysed and
evaluated in both static and dynamic methods. In the end of the
research, two practical examples will be shown for the two methods.
Dental caries is an infectious bacterial disease that results in
localized dissolution and destruction of thecalcified tissues of the
teeth. There are many dental materials which have antibacterial
effects.
The aim of this study is to evaluate th
e antibacterial effects of the
commercial types of Glass Ionomers Cement (GIC)Kaviton® CEM
and Composite Prime Dental® using at the Faculty of Dentistry in
Al- Hawash Private University (HPU), and to make comparsion
between these.