DebasriChakrabarti et al. (2008) [7] gave an algorithm for automatic identification of the verb (V1) + verb (V2) lexical compound verbs (LCpdV) for Hindi language with the success rate of 98% and claimed it to be the first attempt at automatic LCpdV extraction from Hindi corpora. They proposed the heuristic for automatic extraction of LCpdV as, if a stem form of a verb V1 is followed by a verb V2 from a pre-stored list of verbs that can constitute the second component of the LCpdV, then this verb, along with the V2, is considered an LCpdV instance.
MugdhaBapat et al. (2010) [8] proposed a Paradigm-Based Finite State morphological analyser for Marathi language. The morphological analyser takes advantage of finite state machines' efficiency and flexibility in modelling morphotactics, while employing a well-designed set of paradigms to intelligently handle stem alternations by leveraging regularity in inflectional forms and attain the accuracy of 97.18%.
Raj Dabre et al. (2013) [9] proposed a compound word analyser for Marathi with the overall accuracy of 94.83% and described various possible types of compound words that can be formed in Marathi language, which basically follows the rule based technique and is a combination of components like lexicon, suffix replacement rules, inflector and REPO, finite state transducer and morphological parser.
GirishkumarPonkiyaet al. (2018) [10] proposed to paraphrase noun compounds using prepositions by treating noun compounds and their prepositional paraphrases as parallelly aligned word sequences. They used two separate LSTM (Long Short-Term Memory) to encode a noun compound and its prepositional paraphrase, then trained a network to match the encodings of the noun compound and its associated prepositional paraphrase.
Girishkumar Ponkiya et al. (2018) [11] presented a dataset for interpreting noun compounds and used Levi's theory that the noun compounds are formed by either predicate deletion or predicate nominalization [12], to linguistically ground the dataset. They used semantic relation inventory which was based on Frame Net frames and frame elements and conducted these procedures in order to create a standardised dataset with 2,600 samples in the collection. Each noun compound word is annotated with the form of noun compound (predicate deletion vs. predicate nominalization), the frame and frame element through which the noun compound was created in the first place, and its label in three separate datasets.
GirishkumarPonkiya et al. (2020) [13] showed that contextualised language models can be used for unsupervised noun compound paraphrasing and presented two unsupervised techniques for paraphrasing noun compounds: prepositional paraphrasing and free paraphrasing. To produce plausible paraphrases, they used contextualised language models and a feed template and their findings reveal that the unsupervised strategy surpasses supervised approaches for free paraphrasing and are comparable to supervised systems for prepositional paraphrasing.
Jeongin Kim et al. (2021) [21] introduced a Word2VnCR algorithm to replace an OOV word with a semantically related term when an error occurs in morpheme analysis. With the help of this approach, candidate words to be exchanged with the OOV word having the same meaning as OOV are extracted and their semantic similarity to the OOV word's nearby terms can be determined. They also conducted a comparative experiment of Word2VnCR and Word2Vec algorithms and found Word2VnCR to be more efficient in order to replace OOV terms with words that are semantically comparable.
Baby et al. (2022) [22] suggested a post-processing technique that uses the phone cost function to recover context-based OOV words when the list of context words is known. They also looked at acoustic and phonetic similarity-based cost functions, and their research demonstrates that the suggested approach provides OOV context word recovery of more than 50% across several categories.
After going through above-mentioned studies, we found that relatively less work has been done in compound OOV words in Indian languages. A study of Hindi compound verbs (i.e. verb + verb) was presented, as well as diagnostic tests for detecting them and automatic extraction methods from a corpus. A morphological analyser using rule based method for detection and handling of compound words in Marathi has also been developed. However we haven’t found such research for compound OOV words in Hindi language.