This paper addresses the challenges faced by Indian languages in leveraging deep learning for natural language processing (NLP) due to limited resources, annotated datasets, and Transformer-based architectures. We specifically focus on Telugu and aim to construct a Telugu morph analyzer dataset comprising 10,000 sentences. Furthermore, we assess the performance of established multi-lingual Transformer models (m-Bert, XLM-R, IndicBERT) and mono-lingual Transformer models trained from scratch on an extensive Telugu corpus comprising 80,15,588 sentences (BERT-Te). Our findings demonstrate the efficacy of Transformer-based representations pretrained on Telugu data in improving the performance of the Telugu morph analyzer, surpassing existing multi-lingual approaches. This highlights the necessity of developing dedicated corpora, annotated datasets, and machine learning models in a mono-lingual setting. We present benchmark results for the Telugu morph analyzer achieved through simple fine-tuning on our dataset.

Syntactic parsing is the task of assigning a syntactic structure to a sentence. There are two popular syntactic parsing methods: constituency and dependency parsing. Recent works have used syntactic embeddings based on constituency trees, incremental top-down parsing, and other word syntactic features for brain activity prediction given the text stimuli to study how the syntax structure is represented in the brain’s language network. However, the effectiveness of dependency parse trees or the relative predictive power of the various syntax parsers across brain areas, especially for the listening task, is yet unexplored. In this study, we investigate the predictive power of the brain encoding models in three settings: (i) individual performance of the constituency and dependency syntactic parsing based embedding methods, (ii) efficacy of these syntactic parsing based embedding methods when controlling for basic syntactic signals, (iii) relative effectiveness of each of the syntactic embedding methods when controlling for the other. Further, we explore the relative importance of syntactic information (from these syntactic embedding methods) versus semantic information using BERT embeddings. We find that constituency parsers help explain activations in the temporal lobe and middle-frontal gyrus, while dependency parsers better encode syntactic structure in the angular gyrus and posterior cingulate cortex. Although semantic signals from BERT are more effective compared to any of the syntactic features or embedding methods, syntactic embedding methods explain additional variance for a few brain regions.

Transformer-based pretrained models like BERT, GPT-2 and T5 have been finetuned for a large number of natural language processing (NLP) tasks, and have been shown to be very effective. However, while finetuning, what changes across layers in these models with respect to pretrained checkpoints is under-studied. Further, how robust are these models to perturbations in input text? Does the robustness vary depending on the NLP task for which the models have been finetuned? While there exists some work on studying robustness of BERT finetuned for a few NLP tasks, there is no rigorous study which compares this robustness across encoder only, decoder only and encoder-decoder models. In this paper, we characterize changes between pretrained and finetuned language model representations across layers using two metrics: CKA and STIR. Further, we study the robustness of three language models (BERT, GPT-2 and T5) with eight different text perturbations on classification tasks from General Language Understanding Evaluation (GLUE) benchmark, and generation tasks like summarization, free-form generation and question generation. GPT-2 representations are more robust than BERT and T5 across multiple types of input perturbation. Although models exhibit good robustness broadly, dropping nouns, verbs or changing characters are the most impactful. Overall, this study provides valuable insights into perturbation-specific weaknesses of popular Transformer-based models which should be kept in mind when passing inputs.

Named Entity Recognition (NER) is a successful and well-researched problem in English due to the availability of resources. The transformer models, specifically the masked-language models (MLM), have shown remarkable performance in NER during recent times. With growing data in different online platforms, there is a need for NER in other languages too. NER remains to be underexplored in Indian languages due to the lack of resources and tools. Our contributions in this paper include (i) Two annotated NER datasets for the Telugu language in multiple domains: Newswire Dataset (ND) and Medical Dataset (MD), and we combined ND and MD to form Combined Dataset (CD) (ii) Comparison of the finetuned Telugu pretrained transformer models (BERT-Te, RoBERTa-Te, and ELECTRA-Te) with other baseline models (CRF, LSTM-CRF, and BiLSTM-CRF) (iii) Further investigation of the performance of Telugu pretrained transformer models against the multilingual models mBERT, XLM-R, and IndicBERT. We find that pretrained Telugu language models (BERT-Te and RoBERTa) outperform the existing pretrained multilingual and baseline models in NER. On a large dataset (CD) of 38,363 sentences, the BERT-Te achieves a high F1-score of 0.80 (entity-level) and 0.75 (token-level). Further, these pretrained Telugu models have shown state-of-the-art performance on various existing Telugu NER datasets. We open-source our dataset, pretrained models, and code.

Several popular Transformer based language models have been found to be successful for text-driven brain encoding. However, existing literature leverages only pretrained text Transformer models and has not explored the efficacy of task-specific learned Transformer representations. In this work, we explore transfer learning from representations learned for ten popular natural language processing tasks (two syntactic and eight semantic) for predicting brain responses from two diverse datasets: Pereira (subjects reading sentences from paragraphs) and Narratives (subjects listening to the spoken stories). Encoding models based on task features are used to predict activity in different regions across the whole brain. Features from coreference resolution, NER, and shallow syntax parsing explain greater variance for the reading activity. On the other hand, for the listening activity, tasks such as paraphrase generation, summarization, and natural language inference show better encoding performance. Experiments across all 10 task representations provide the following cognitive insights: (i) language left hemisphere has higher predictive brain activity versus language right hemisphere, (ii) posterior medial cortex, temporo-parieto-occipital junction, dorsal frontal lobe have higher correlation versus early auditory and auditory association cortex, (iii) syntactic and semantic tasks display a good predictive performance across brain regions for reading and listening stimuli resp.