Context-Aware Bidirectional Neural Model for Sindhi Named Entity Recognition

Abstract

Named entity recognition (NER) is a fundamental task in many natural language processing (NLP) applications, such as text summarization and semantic information retrieval. Recently, deep neural networks (NNs) with the attention mechanism yield excellent performance in NER by taking advantage of character-level and word-level representation learning. In this paper, we propose a deep context-aware bidirectional long short-term memory (CaBiLSTM) model for the Sindhi NER task. The model relies upon contextual representation learning (CRL), bidirectional encoder, self-attention, and sequential conditional random field (CRF). The CaBiLSTM model incorporates task-oriented CRL based on joint character-level and word-level representations. It takes character-level input to learn the character representations. Afterwards, the character representations are transformed into word features, and the bidirectional encoder learns the word representations. The output of the final encoder is fed into the self-attention through a hidden layer before decoding. Finally, we employ the CRF for the prediction of label sequences. The baselines and the proposed CaBiLSTM model are compared by exploiting pretrained Sindhi GloVe (SdGloVe), Sindhi fastText (SdfastText), task-oriented, and CRL-based word representations on the recently proposed SiNER dataset. Our proposed CaBiLSTM model achieved a high F1-score of 91.25% on the SiNER dataset with CRL without relying on additional handmade features, such as hand-crafted rules, gazetteers, or dictionaries.

State-of-art approach for Indian Language based on NER: Comprehensive Review

“…For this research, 15k-sentences from the Punjabi data corpus and many newspapers were used, and they were annotated utilizing the open-source annotation tool Doccano. Ali et al [83] discussed pretrained Sindhi Glove (SdGlove), Sindhi fastText (Sdfast Text), task-oriented and CRL-based word representations on the recently presented SiNER dataset, the baselines and CaBiLSTM model are compared which has been proposed. After comparison of these two models, it achieved a high F1-score of 91.25% by using CaBiLSTM on the SiNER dataset with CRL.…”

“…The authors applied Bi-LSTM [85] based on DL which contains POS and CNN embedding character and overall accuracy has been calculated by using F1-score, recall and precision. Ali et al [2], [94] introduced a NER dataset for the limited resources Sindhi language with quality benchmarks with SiNER. It has 1,338 updates along with more than 1.35 million tokens that were gathered with the begin-inside-outside (BIO) tagging method used by Kawish and Awami Awaz Sindhi newspapers as the suggested dataset has a good potential of being a useful tool for statistical Sindhi language processing.…”

“…The system produced results utilizing hybrid unigrams and bigrams on the IJCNLP NE corpus and the CRL NE corpus of 92.65 and 87.6% and 92.47 and 86.83%, respectively. Ali et al [21] described the present challenges in developing the SNER system while taking the need for language processing tools into consideration as the purpose of this essay is to explore challenges and potential areas for further study in the subject of NER in the Sindhi language. Few existing works are discussed in table 6 related to hybrid approach (i.e., combination of any other two methods like ML, DL and rule-based approach).…”

“…Litake et al [10] Hindi, Marathi IJCNLP 80% Mehta et al [102] Hindi, Indo-aryan Large Corpus 90.7% Saha et al [103] Hindi Test Corpus 73.87% Pant et al [47] Kumauni Annotated data Corpus 85% Krishnan et al [104] Malayalam Wikipedia, Google Graph F1=78% Liu et al [105] Urdu MULTIPLE, IDENTIFIER 80% Srivastava et al [106] Hindi IJCNLP-08 P=96.05, F=91.25%, R=86.90% Naz et al [107] Urdu IJCNLP and ACL corpus F=92.47%, R=95.57%, P=89.57% Ali et al [21] Sindhi IJCNLP-2008 82%…”

“…NER for Sindhi language would involve building a system that can accurately identify and classify named entities within Sindhi text. "It's worth mentioning that NER for low-resource languages like Sindhi can be challenging due to the limited availability of annotated data and linguistic resources [2], [9], [16], [21], [22]. Additionally, the accuracy of an NER system depends on the quality and coverage of the training data, as well as the complexity of the language and the types of entities we want to recognize".…”