Combining Curriculum Learning and Knowledge Distillation for Dialogue Generation

Curriculum learning, a machine training strategy that feeds training instances to the model from easy to hard, has been proven to facilitate the dialogue generation task. Meanwhile, knowledge distillation, a knowledge transformation methodology among teachers and students networks can yield significant performance boost for student models. Hence, in this paper, we introduce a combination of curriculum learning and knowledge distillation for efficient dialogue generation models, where curriculum learning can help knowledge distillation from data and model aspects. To start with, from the data aspect, we cluster the training cases according to their complexity, which is calculated by various types of features such as sentence length and coherence between dialog pairs. Furthermore, we employ an adversarial training strategy to identify the complexity of cases from model level. The intuition is that, if a discriminator can tell the generated response is from the teacher or the student, then the case is difficult that the student model has not adapted to yet. Finally, we use self-paced learning, which is an extension to curriculum learning to assign weights for distillation. In conclusion, we arrange a hierarchical curriculum based on the above two aspects for the student model under the guidance from the teacher model. Experimental results demonstrate that our methods achieve improvements compared with competitive baselines.

Combining sentence and table evidence to predict veracity of factual claims using TaPaS and RoBERTa

This paper describes a method for retrieving evidence and predicting the veracity of factual claims, on the FEVEROUS dataset. The evidence consists of both sentences and table cells. The proposed method is part of the FEVER shared task. It uses similarity scores between TF-IDF vectors to retrieve the textual evidence and similarity scores between dense vectors created by fine-tuned TaPaS models for tabular evidence retrieval. The evidence is passed through a dense neural network to produce a veracity label. The FEVEROUS score for the proposed system is 0.126.

Event Prominence Extraction Combining a Knowledge-Based Syntactic Parser and a BERT Classifier for Dutch

A core task in information extraction is event detection that identifies event triggers in sentences that are typically classified into event types. In this study an event is considered as the unit to measure diversity and similarity in news articles in the framework of a news recommendation system. Current typology-based event detection approaches fail to handle the variety of events expressed in real-world situations. To overcome this, we aim to perform event salience classification and explore whether a transformer model is capable of classifying new information into less and more general prominence classes. After comparing a Support Vector Machine (SVM) baseline and our transformer-based classifier performances on several event span formats, we conceived multi-word event spans as syntactic clauses. Those are fed into our prominence classifier which is fine-tuned on pre-trained Dutch BERT word embeddings. On top of that we outperform a pipeline of a Conditional Random Field (CRF) approach to event-trigger word detection and the BERT-based classifier. To the best of our knowledge we present the first event extraction approach that combines an expert-based syntactic parser with a transformer-based classifier for Dutch.

YNU-HPCC at SemEval-2021 Task 6: Combining ALBERT and Text-CNN for Persuasion Detection in Texts and Images

In recent years, memes combining image and text have been widely used in social media, and memes are one of the most popular types of content used in online disinformation campaigns. In this paper, our study on the detection of persuasion techniques in texts and images in SemEval-2021 Task 6 is summarized. For propaganda technology detection in text, we propose a combination model of both ALBERT and Text CNN for text classification, as well as a BERT-based multi-task sequence labeling model for propaganda technology coverage span detection. For the meme classification task involved in text understanding and visual feature extraction, we designed a parallel channel model divided into text and image channels. Our method achieved a good performance on subtasks 1 and 3. The micro F1-scores of 0.492, 0.091, and 0.446 achieved on the test sets of the three subtasks ranked 12th, 7th, and 11th, respectively, and all are higher than the baseline model.

Combining ability analysis for seed yield and some yield compenents traits in six lines of chickpea (Cicer arietinum L.)

A half diallel making design was used to determine combining ability and heterosis of six chickpea lines and their hybrid combinations with R.C.B.D. in three replication for seed yield per plant, number of pods per branch, pods number per plant, first pod height and 100- seed weight, during 2015 and 2016 seasons, at the research Center of Al-Gab (G.C.S.A.R). Genotypes, general (GCA) and specific (SCA) combining ability mean squares were significant for all studied traits. The ratio σ2 GCA / σ2 SCA were detected for all traits and showed that non- additive gene action was more important than additive gene action in controlling all studied traits. GCA effects showed that the lines الإسباني was good general combiner for traits:seed yield per plant, pods per plant, 100- seed weight, days to maturity, protein per cent. SCA effects showed that(Algerian×Spanish) hybrid was the good specific combiner for seed yield per plant, pods per plant, 100- seed weight, harvest index. On the other side(Algerian×Spanish) was showed significant desirable heterosis values for seed yield per plant and 100- seed weight, days to maturity, harvest index.

Genetic analysis for yield and its components in hybrids of chickpea (Cicer arietinum L.)

A half diallel making design was used to determine combining ability and heterosis of six chickpea lines and their hybrid combinations with R.C.B.D. in three replication for seed yield per plant, number of pods per branch, pods number per plant, first pod height and 100- seed weight, during 2015 and 2016 seasons, at the research Center of Al-Gab (G.C.S.A.R). Genotypes, general (GCA) and specific (SCA) combining ability mean squares were significant for all studied traits. The ratio σ2 GCA / σ2 SCA were detected for all traits and showed that non- additive gene action was more important than additive gene action in controlling all studied traits. GCA effects showed that the lines P2(IL.10158) was good general combiner for number of pods per branch and plant, also, 100- seed weight where, P3 (IL.5883) and P4(IL.4) were good general combiner for first pod height and seed yield per plant, respectively. SCA effects showed that(P1×P2) hybrid was the best F1 crosses combination for seed yield per plant. On the other side (P1×P2), (P1×P4) and (P2×P4) were showed positive and significant heterosis values for seed yield per plant and 100- seed weight also first pod height and number of pod per branch and plant respectively.

General and Specific Combining Ability and Genetic Parameters (Zea mays L.) of Single Crosses in Corn

Six inbred lines of maize namely; A (1), B (2), C (3), D (4), E (5) and F (6) were used in half diallel cross. The seeds of inbred lines and its single cross hybrids were cultivated in an experiment using randomized completely block design (RCBD) with three replicates, at Twaitha Research Station, Plant Breeding Improvement Center, Iraq, during autumn season (2016). The parents and F1 were significantly differed at 5% for number of days to tasseling and silking, plant height (cm), ear diameter (cm) and yield per plant. Some hybrids showed a significant desirable heterosis for studied traits such as days to tasseling and silking. While plant height and yield per plant for all hybrids showed a significant desirable heterosis (deviation of F1 from mid parents). The mean squares of general and specific combining ability were highly significant for all traits. The additive and dominance variances were differed from zero for all studied traits. The broad sense heritability values were high for all studied traits. Narrow sense heritability was moderate for number of days to tasseling, plant height and ear diameter, but it was low for number of days to silking and yield per plant. The average degree of dominance was higher than one for all traits.

Combining Ability and Heritability of Some Productivity Characters in Some Hybrids of Cotton (Gossypium hirsutum L.)

The research was carried out at Tal Hedya Research Center in Aleppo, General Commission for Agriculture Scientific Research (GCSAR), Syria, during 2011 and 2012 seasons. Seven genotypes of cotton were used, and complete diallel hybridization was made to study general and specific combining ability of some productivity traits (sympodial branch number, actual boll number and seed cotton yield). General combining ability (GCA) indicated that the parental genotype Cherpan432 had a high significant GCA for sympodial branch number and actual boll number, but the parents Aleppo118 and Deir El-Zour22 had the high GCA for seed cotton yield. This is a clear indication that these parental genotypes had the largest number of additive genes action, which plays important role in the inheritance of the above-mentioned traits. The estimation of SCA values showed favorite and high significant values in many hybrids resulted from parents, which had highly significant GCA, which means that the gene action type is (additive x additive), and this refers that these hybrids possessing the largest number of additive genes. High heritability broad sense refers to importance of genetic variance in the inheritance of all characters, but heritability in narrow sense values were low in general, indicating the importance of dominance and epistasis genes in the inheritance. According to this result it is recommended to follow the cross method for improvement of the studied traits.

Combining between rhetoric and poetry at the Gahleon and its impact on poetry

This research deals with the phenomenon of combining rhetoric and poetry at a number of the Gahleon, and the statement of the effect of this combining in their poetry, explaining through the issue of conflict of prestige betweenthe orator and the poet in pre-Islamic era, and how this conflict has been compromised between who combined the technocracy, as the research shows the reason oflack of those who gathered between technocracy. The effect of combining in poetry is dealt with research on two levels: the stylistic level, and shows all forms of repetition: repeating letters, words and methods. While appears in the substantive level (moral): the pursuit of incomprehensible, the concentration on thinking and subjects, and the large number of wisdom and wills.

Estimate Some Genetic Parameters for Grain Yield and Some of Its Contributing Traits in Bread Wheat Crosses

The Experiment was conducted during the successive growing seasons (2013/2014- 2014/2015) where half–diallel hybridization was conducted between eight genotypes of bread wheat.