Exploring Syntactic Patterns in Urdu: A Deep Dive into Dependency Analysis

Parsing is the process of breaking a sentence into its grammatical components and identifying the syntactic structure of the sentence. The syntactically correct sentence structure is achieved by assigning grammatical labels to its constituents using lexicon and syntactic rules. In linguistics, parser is extremely useful due to the number of different applications like name entity recognition, QA systems and information extraction, etc. The two most common techniques used for parsing are phrase structure and dependency Structure. Because Urdu is a low-resource language, there has been little progress in building an Urdu parser. A comparison of several parsers revealed that the dependency parsing approach is better suited for order-free languages such as Urdu. We have made significant progress in parsing Urdu, a South Asian language with a complex morphology. For Urdu dependency parsing, a basic feature model consisting of word location, word head, and dependency relation is employed as a starting point, followed by more complex feature models. The dependency tagset is designed after careful consideration of the complex morphological structure of the Urdu language, word order variation, and lexical ambiguity and it contains 22 tags. Our dataset comprises of sentences from news articles, and we tried to include sentences of different complexity (which is quite challenging), to get reliable results. All experiments are performed using MaltParser, exploring all 9 algorithms and classifiers. We have achieved a 70 percent overall best-labeled accuracy (LA), as well as an 84 percent overall best-unlabeled attachment score (UAS) using the Nivreeager algorithm. The comparison of output data with treebank test data that has been manually parsed is then used to carry out error assessment and to identify the errors produced by the parser.

On the Limitations of Large Language Models : False Attribution

In this work, we provide insight into one important limitation of large language models (LLMs), i.e. false attribution, and introduce a new hallucination metric - Simple Hallucination Index (SHI). The task of automatic author attribution for relatively small chunks of text is an important NLP task but can be challenging. We empirically evaluate the power of 3 open SotA LLMs in zero-shot setting (LLaMA-2-13B, Mixtral 8x7B, and Gemma-7B), especially as human annotation can be costly. We collected the top 10 most popular books, according to Project Gutenberg, divided each one into equal chunks of 400 words, and asked each LLM to predict the author. We then randomly sampled 162 chunks for human evaluation from each of the annotated books, based on the error margin of 7% and a confidence level of 95% for the book with the most chunks (Great Expectations by Charles Dickens, having 922 chunks). The average results show that Mixtral 8x7B has the highest prediction accuracy, the lowest SHI, and a Pearson's correlation (r) of 0.737, 0.249, and -0.9996, respectively, followed by LLaMA-2-13B and Gemma-7B. However, Mixtral 8x7B suffers from high hallucinations for 3 books, rising as high as an SHI of 0.87 (in the range 0-1, where 1 is the worst). The strong negative correlation of accuracy and SHI, given by r, demonstrates the fidelity of the new hallucination metric, which is generalizable to other tasks. We publicly release the annotated chunks of data and our codes to aid the reproducibility and evaluation of other models.

Instruction Makes a Difference

We introduce Instruction Document Visual Question Answering (iDocVQA) dataset and Large Language Document (LLaDoc) model, for training Language-Vision (LV) models for document analysis and predictions on document images, respectively. Usually, deep neural networks for the DocVQA task are trained on datasets lacking instructions. We show that using instruction-following datasets improves performance. We compare performance across document-related datasets using the recent state-of-the-art (SotA) Large Language and Vision Assistant (LLaVA)1.5 as the base model. We also evaluate the performance of the derived models for object hallucination using the Polling-based Object Probing Evaluation (POPE) dataset. The results show that instruction-tuning performance ranges from 11X to 32X of zero-shot performance and from 0.1% to 4.2% over non-instruction (traditional task) finetuning. Despite the gains, these still fall short of human performance (94.36%), implying there's much room for improvement.