Jaspreet Singh

Jonas Wallat, Jaspreet Singh, Avishek Anand

Probing complex language models has recently revealed several insights into linguistic and semantic patterns found in the learned representations. In this article, we probe BERT specifically to understand and measure the relational knowledge it captures in its parametric memory. While probing for linguistic understanding is commonly applied to all layers of BERT as well as fine-tuned models, this has not been done for factual knowledge. We utilize existing knowledge base completion tasks (LAMA) to probe every layer of pre-trained as well as fine-tuned BERT models(ranking, question answering, NER). Our findings show that knowledge is not just contained in BERT's final layers. Intermediate layers contribute a significant amount (17-60%) to the total knowledge found. Probing intermediate layers also reveals how different types of knowledge emerge at varying rates. When BERT is fine-tuned, relational knowledge is forgotten. The extent of forgetting is impacted by the fine-tuning objective and the training data. We found that ranking models forget the least and retain more knowledge in their final layer compared to masked language modeling and question-answering. However, masked language modeling performed the best at acquiring new knowledge from the training data. When it comes to learning facts, we found that capacity and fact density are key factors. We hope this initial work will spur further research into understanding the parametric memory of language models and the effect of training objectives on factual knowledge. The code to repeat the experiments is publicly available on GitHub.

Jonas Wallat, Jaspreet Singh, Avishek Anand

Probing complex language models has recently revealed several insights into linguistic and semantic patterns found in the learned representations. In this paper, we probe BERT specifically to understand and measure the relational knowledge it captures. We utilize knowledge base completion tasks to probe every layer of pre-trained as well as fine-tuned BERT (ranking, question answering, NER). Our findings show that knowledge is not just contained in BERT's final layers. Intermediate layers contribute a significant amount (17-60%) to the total knowledge found. Probing intermediate layers also reveals how different types of knowledge emerge at varying rates. When BERT is fine-tuned, relational knowledge is forgotten but the extent of forgetting is impacted by the fine-tuning objective but not the size of the dataset. We found that ranking models forget the least and retain more knowledge in their final layer. We release our code on github to repeat the experiments.

Gaurav Kumar, Rishabh Joshi, Jaspreet Singh et al.

The problem of building a coherent and non-monotonous conversational agent with proper discourse and coverage is still an area of open research. Current architectures only take care of semantic and contextual information for a given query and fail to completely account for syntactic and external knowledge which are crucial for generating responses in a chit-chat system. To overcome this problem, we propose an end to end multi-stream deep learning architecture which learns unified embeddings for query-response pairs by leveraging contextual information from memory networks and syntactic information by incorporating Graph Convolution Networks (GCN) over their dependency parse. A stream of this network also utilizes transfer learning by pre-training a bidirectional transformer to extract semantic representation for each input sentence and incorporates external knowledge through the the neighborhood of the entities from a Knowledge Base (KB). We benchmark these embeddings on next sentence prediction task and significantly improve upon the existing techniques. Furthermore, we use AMUSED to represent query and responses along with its context to develop a retrieval based conversational agent which has been validated by expert linguists to have comprehensive engagement with humans.

Jaspreet Singh, Johannes Hoffart, Avishek Anand

Linking entities like people, organizations, books, music groups and their songs in text to knowledge bases (KBs) is a fundamental task for many downstream search and mining applications. Achieving high disambiguation accuracy crucially depends on a rich and holistic representation of the entities in the KB. For popular entities, such a representation can be easily mined from Wikipedia, and many current entity disambiguation and linking methods make use of this fact. However, Wikipedia does not contain long-tail entities that only few people are interested in, and also at times lags behind until newly emerging entities are added. For such entities, mining a suitable representation in a fully automated fashion is very difficult, resulting in poor linking accuracy. What can automatically be mined, though, is a high-quality representation given the context of a new entity occurring in any text. Due to the lack of knowledge about the entity, no method can retrieve these occurrences automatically with high precision, resulting in a chicken-egg problem. To address this, our approach automatically generates candidate occurrences of entities, prompting the user for feedback to decide if the occurrence refers to the actual entity in question. This feedback gradually improves the knowledge and allows our methods to provide better candidate suggestions to keep the user engaged. We propose novel human-in-the-loop retrieval methods for generating candidates based on gradient interleaving of diversification and textual relevance approaches. We conducted extensive experiments on the FACC dataset, showing that our approaches convincingly outperform carefully selected baselines in both intrinsic and extrinsic measures while keeping users engaged.

Jaspreet Singh, Zeon Trevor Fernando, Saniya Chawla

In addition to user-generated content, Open Educational Resources are increasingly made available on the Web by several institutions and organizations with the aim of being re-used. Nevertheless, it is still difficult for users to find appropriate resources for specific learning scenarios among the vast amount offered on the Web. Our goal is to give users the opportunity to search for authentic resources from the Web and reuse them in a learning context. The LearnWeb-OER platform enhances collaborative searching and sharing of educational resources providing specific means and facilities for education. In the following, we provide a description of the functionalities that support users in collaboratively collecting, selecting, annotating and discussing search results and learning resources.