hindi word problem solving

Verb Categorisation for Hindi Word Problem Solving

Word problem Solving is a challenging NLP task that deals with solving mathematical problems described in natural language. Recently, there has been renewed interest in developing word problem solvers for Indian languages. As part of this paper, we have built a Hindi arithmetic word problem solver which makes use of verbs. Additionally, we have created verb categorization data for Hindi. Verbs are very important for solving word problems with addition/subtraction operations as they help us identify the set of operations required to solve the word problems. We propose a rule-based solver that uses verb categorisation to identify operations in a word problem and generate answers for it. To perform verb categorisation, we explore several approaches and present a comparative study.

Harshita Sharma, Pruthwik Mishra, Dipti Misra Sharma IIIT-Hyderabad {harshita.sharma, pruthwik.mishra}@research.iiit.ac.in, [email protected]

1 Introduction

Verb Categorisation is the most intuitive and explainable semantic parsing approach for word problem solving. This approach was introduced in Hosseini et al. ( 2014 ) . It uses verbs to identify operations required to solve a word problem. The idea is to identify the following parts of a word problem on top of which we use verb categories to perform calculations:

Entities : Objects whose quantity is observed or updated through the course of the word problem.

Attributes of Entities : A characteristic quality or feature of an entity. These are usually marked by adjectives.

Containers : A container refers to a group of entities. It may refer to any animate/inanimate object that possesses or contains entities.

Quantities of Entities : The number of entities in a given container. Quantities in a container can be known or unknown.

This can be explained using the following example:

Leveraging the insights provided by Hosseini et al. ( 2014 ) , as part of this paper, we make the following contributions:

Redefine verb categories for word problem solving.

Create verb categorisation data for the Hindi language.

Introduce three new verb categorisation approaches and provide a comprehensive comparative analysis of these approaches.

Introduce a rule-based solver 1 1 1 Code and data can be found here: https://github.com/hellomasaya/verb-cat-for-hindi-wps that uses verbs to identify specific mathematical operations to solve word problems.

2 Motivation

Let us take an example to understand the vital role of verbs in solving a word problem as shown in the following figures. Figure  2 shows a word problem with containers, entities, and their quantities by masking the verbs. Here, we cannot identify the operation needed to reach the final state and answer the question asked in the word problem. However, in Figure 3 , we are able to reach the final state. Moreover, when we change the verb from Figure  3 to Figure  4 , the operation also changes.

Verb Category	Definition
Observation	It states just the presence of entities in a container
Positive	It states the number of entities being added to a container or which are created in a container.
Negative	It states the quantity of entities being removed or destroyed from a container.
Positive Transfer	It is associated with statements that involve two containers. It states a transfer of the number of entities from the second container to the first.
Negative Transfer	It is associated with statements that involve two containers. It states a transfer of the quantity of entities from the first container to the second.

3 Verb Categorisation

This section focuses on the first step of word problem solving using verb categorisation, i.e. classifying verbs into semantic categories. Since verbs can be used to identify only positive and negative operations, we filtered the HAWP dataset 2 2 2 https://github.com/hellomasaya/hawp Sharma et al. ( 2022 ) to have only word problems involving addition and subtraction operations. Verbs tell us whether entities are observed, created, destroyed, or transferred. For multiplication and division, we need another layer of categorisation with different Part of Speech categories on top of verb categorisation.

3.1 Verb Categories

Table  1 lists the five categories we have used in this paper. We also included a sixth category - na . During POS tagging, non-verbs were tagged as verbs; these tokens were put into the na category.

3.2 Annotation of Verbs

In the HAWP dataset (2336 word problems), 1713 word problems are based on addition and subtraction operations. These problems feature in our dataset for word problem solving using verb categorisation. In these 1713 word problems, there are around 200 unique verbs. These verbs were annotated with the categories mentioned above.

Hosseini et al. ( 2014 ) have seven verb categories that are container-centric. They have two additional categories of Construct and Destroy apart from the ones defined in Table  1 . But these two resemble Positive and Negative categories, respectively. Hence, we decide to drop these two categories.

For the verb annotation task, two annotators with post-graduate levels of education in computational linguistics are involved. We conduct experiments to evaluate the inter-annotator agreement between them on 225 verbs from 100 sentences. The Fleiss’ 3 3 3 https://en.wikipedia.org/wiki/Fleiss%27_kappa kappa score of agreement is 0.89, which denotes almost perfect agreement. There was maximum disagreement between Observation and Positive classes.

3.3 Approaches

We tried mainly three kinds of approaches, which are detailed in the following subsections. All the approaches are evaluated using a 5-fold cross-validation technique.

3.3.1 Verb Distance

The first approach is the training less method using verb distance. Each verb in Hindi is represented by its pre-trained FastText word vector Grave et al. ( 2018 ) of 300 dimensions. A test verb is assigned the verb category corresponding to its closest training verb. We implemented this approach using the 1-nearest neighbour approach.

3.3.2 Statistical Models

3.3.2.1   data preparation.

The idea is to use a bag of words representation for a verb and its neighbours in their actual order as a sample and the category of the verb as the label. We created samples for the task using word-level information as indicated in Figure  6 . After trying context windows of different sizes, we finalised the context window size as 7.

Therefore, we will have word-level information for three neighbours to the right of the verb and the same for three neighbours to the left of the verb. We parse each sentence using an in-house shallow parser Mishra et al. ( 2023 ) for identifying the POS tag and root of each word. We used ISC-parser from Natural language tool-kit for Indian Language Processing 4 4 4 https://github.com/iscnlp/iscnlp/tree/master/iscnlp to get dependency tags of each token in each sentence of each word problem in the dataset.

3.3.2.2   Experimental Setup

We performed this classification task using 3 machine-learning approaches:

Logistic Regression

Random Forest

Support Vector Machines (SVM)

All these models have been implemented using Scikit-learn Pedregosa et al. ( 2011 ) machine learning framework.

3.3.3 MuRIL Contextual Embeddings

Contextual embeddings, especially BERT Devlin et al. ( 2019 ) based embeddings, have been shown to be very effective for classification and generalization tasks. BERT is trained in two stages: pre-training and fine-tuning. The model is first trained on a huge monolingual corpus to learn language-specific representations and then fine-tuned on a downstream task. In our case, the downstream task is the verb categorization task. As this is a text or sentence classification task, it is a perfect test for using BERT or BERT-like models. For this, we used MuRIL Khanuja et al. ( 2021 ) , a multilingual transformer Vaswani et al. ( 2017 ) model trained on English and 16 Indian languages. MuRIL is pre-trained using masked language modelling as well as translation language modelling. It has a combined vocabulary of 197K words.

3.3.3.1   Data Preparation

We used the 1713 word problems from HAWP. Since MuRIL can handle large contexts, we do not limit ourselves to a fixed context window. For this task, all the words till a verb is encountered constitute a sample. A total of 6506 samples were created for verb categorization. Let us take an example to understand this better.

Original Question: kanishk ko samudr tat par 47 seepiyaan mileen, usane laila ko 25 seepiyaan deen. usake paas ab kitanee seepiyaan hain?

Gloss: Kanishk found 47 shells on the beach, he gave 25 shells to Laila. How many shells does he have now?

Samples for Verb Categorization

kanishk ko samudr tat par 47 seepiyaan mileen

Gloss: Kanishk (found) 47 shells on the beach

kanishk ko samudr tat par 47 seepiyaan mileen, usane laila ko 25 seepiyaan deen .

Gloss: Kanishk found 47 shells on the beach, he (gave) 25 shells to Laila.

usake paas ab kitanee seepiyaan hain ?

Gloss: How many shells does he (have) now?

3.3.3.2   Experimental Setup

MuRIL has 236 million parameters, and it uses AdamW Loshchilov and Hutter ( 2017 ) optimizer. We use a 5-fold cross-validation technique to evaluate the model. MuRIL is fine-tuned for ten epochs with a batch size of 4. MuRIL-based text classification model is implemented using HuggingFace Wolf et al. ( 2019 ) library.

3.4 Results and Discussion

The results from all models are shown in Table  2 .

Approach	F1-score
Verb Distance	0.895
Logistic Regression	0.865
Random Forest	0.883
Support Vector Machines	0.904
MuRIL Fine-tuning	0.962

We can observe that MuRIL Fine-tuning outperforms other approaches by a significant margin. The class na contains the highest classification error. The major cause of ambiguity is between the Observation and Positive in all the models.

We build a simple rule-based system that takes in a word problem and generates answers to the word problem. For each problem, we iterate through all tokens in each of its sentences and follow the rules mentioned below.

4.1 Find container, quantity and entity

A container is a proper noun or adverb of place .

A quantity is a number . Whenever a quantity is found, the last identified container is associated with this quantity.

An entity is a noun . If there is an adjective associated with this entity, it is clubbed with the entity. When a word problem has the Rupee symbol, the entity is taken to be this symbol itself. Examples of these rules can be found in Figures  11 and 10 .

4.2 Store States

Here, a state refers to the status of an entity that stores information about an entity, its container, its associated quantity, and any attributes of the entity.

Once an entity is found, the associated quantity and container are used to form a state.

Before storing the quantity in a state, if the verb that follows the identified entity of this state has its verb category as Negative, the quantity is negated and stored. A detailed example of applying these rules can be found in Figure  10 .

4.3 Handle Transfer Category

Once a verb is found in the word problem, we check for Transfer categories. We check if the verb belongs to the Positive Transfer or the Negative Transfer category from our verb categorisation exercise.

If the Transfer verb category is found, we find transfer components, i.e. transfer containers (two containers b/w which transfer is taking place) and the quantity of entities being transferred.

Then, we iterate through the states and find which already present states have transfer-container 1 and transfer-container 2. Then, we check if transfer-entity is present in these states.

Finally, we update the previous states based on which containers and entities are already available in the previous states. These cases for positive and negative categories can be seen in Fig 8 and Fig 9 . A detailed example of applying these rules can be found in Fig 11 .

4.4 Finding Answer

Find Question Entity (and Question Container, only in the case when a transfer verb is encountered) from the question using the same rules mentioned in Section 4.1 .

Find Main Operation.

If a transfer verb category is encountered in the word problem, the main operation is Transfer.

If any positive indicator is present in question, the main operation is Positive.

If any negative indicator is present in question, the main operation is negative.

The main operation is positive if none of the above conditions are met.

Indicators	Examples
Positive Indicators	‘kul’, ‘milakar’, ’milkar’ etc.
Negative Indicators	‘mukable’, ‘tulna’, ‘pehle’, ‘chahiye’ etc.

If the main operation is Transfer, our calculation is already complete as part of 4.3 . We find the state that has the answer to the question by looking at all the states we created, and whichever state matches the question’s container and entity pair, we return its quantity as the answer. A detailed example of the transfer verb category is explained in Fig 11 . More examples can be found in the Appendix.

If the main operation is Negative, we find all states that have the same entity as the question entity. Then, keeping the quantity in the first state as it is, we subtract the quantities of the states that follow from it to finally reach the answer.

If the main operation is Positive, we find all states that have the same entity as the question entity and add all the quantities of these states to finally reach the answer. A detailed example of this case is explained in Fig 10 .

4.5 More Rules

If the final answer calculated by the solver is negative, we return its absolute value.

While identifying relevant entities from states, if the entities from the question and a state match but the attribute is missing in either state or question, we still regard the entity as relevant.

If the entity in the word problem is found to be one of ‘paisa’, ‘keemat’, ‘laagat’, and ‘rupay’, we change it to the Rupee symbol.

If a quantity is found without an entity or container, we retain the same entity and container from the last state and create a new state with the quantity found. This is called circumscription assumption McCarthy ( 1980 ) .

If the entity in question is not found in states, we assume the entity of the first state to be the entity in the question and perform the steps for finding the answer.

Detailed examples of all rules can be found in the Appendix.

4.6 Results and Discussion

The solver was tested on test sets using predicted verb categories, and an average accuracy of 41.2% was reported, which is comparable to 40.04% reported by Sharma et al. ( 2022 ) for one-operation problems in the HAWP dataset. Some of the cases in which the solver fails are listed below:

Irrelevant Information: The solver fails to identify some cases of irrelevant information.

Error in entity/container/action identification.

Set Completion: The solver fails to handle word problems which require the knowledge of set completion.

Parsing Errors: Errors caused by incorrectly tagged part of speech. This also includes cases when parsers miss foreign words.

Rules: There are cases when a rule that works for some examples may not work for others.

Examples of these cases can be found in the Appendix in Table 4 .

5 Limitations

Apart from the limitations of the solver, the method of using verb categorisation to solve word problems also has some limitations. As stated, solving word problems using verb categorisation is only limited to addition and subtraction word problems because verbs can only help us identify these operations. Moreover, there were errors in the dependency labels. Since verb categorisation very heavily relies on these parsers for finding verb categories and identifying entities, containers and actions/verbs for solving word problems. This adds to the limitation of this method.

6 Conclusion and Future Work

In this paper, we create a rule-based and easily explainable solver that uses a verb categorisation technique to identify operations to solve word problems. This can be used as a teaching aid for both students and teachers. As part of the verb categorisation task, we run experiments with three approaches: Verb Distance (no training involved), statistical, and neural approaches using MuRIL. As part of future work, we will explore more approaches to improve the accuracy of our solver and its range, i.e., solving word problems with multiplication and division operations.

• Devlin et al. (2019) Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding . In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages 4171–4186, Minneapolis, Minnesota. Association for Computational Linguistics.
• Grave et al. (2018) Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Armand Joulin, and Tomas Mikolov. 2018. Learning word vectors for 157 languages . In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018) , Miyazaki, Japan. European Language Resources Association (ELRA).
• Hosseini et al. (2014) Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. 2014. Learning to solve arithmetic word problems with verb categorization . In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 523–533, Doha, Qatar. Association for Computational Linguistics.
• Khanuja et al. (2021) Simran Khanuja, Diksha Bansal, Sarvesh Mehtani, Savya Khosla, Atreyee Dey, Balaji Gopalan, Dilip Kumar Margam, Pooja Aggarwal, Rajiv Teja Nagipogu, Shachi Dave, et al. 2021. Muril: Multilingual representations for indian languages . arXiv preprint arXiv:2103.10730 .
• Loshchilov and Hutter (2017) Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization . arXiv preprint arXiv:1711.05101 .
• McCarthy (1980) John McCarthy. 1980. Circumscription—a form of non-monotonic reasoning . Artificial intelligence , 13(1-2):27–39.
• Mishra et al. (2023) Pruthwik Mishra, Vandan Mujadia, and Dipti Misra Sharma. 2023. Multi Task Learning Based Shallow Parsing for Indian Languages.
• Pedregosa et al. (2011) F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine learning in Python . Journal of Machine Learning Research , 12:2825–2830.
• Sharma et al. (2022) Harshita Sharma, Pruthwik Mishra, and Dipti Sharma. 2022. HAWP: a dataset for Hindi arithmetic word problem solving . In Proceedings of the Thirteenth Language Resources and Evaluation Conference , pages 3479–3490, Marseille, France. European Language Resources Association.
• Vaswani et al. (2017) Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need . Advances in neural information processing systems , 30.
• Wolf et al. (2019) Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. 2019. Huggingface’s transformers: State-of-the-art natural language processing . arXiv preprint arXiv:1910.03771 .

Appendix A Appendix

A.1 examples of rules used in our hindi word problem solver, a.1.1 example when main operation is transfer.

As mentioned in Section 4.4 , the main operation is ‘transfer’ when a transfer verb category is encountered in the word problem. The transfer verb category may have two types: Positive and negative transfer. An example of Negative Transfer is covered as part of Figure 11 . Figure 12 illustrates an example of how a problem with positive transfer verb category is solved by the solver.

A.1.2 Examples when the Main operation is Negative

Section 4.4 states that the main operation is Negative when no transfer verb category is found, and a negative indicator is present in question. Figure 13 states an example of the same.

A.1.3 Examples of rules mentioned in Section 4.5

If the final answer calculated by the solver is negative, we return its absolute value. Example in Figure 14 .

If the entity in the word problem is found to be one of ‘paisa’, ‘keemat’, ‘laagat’, and ‘rupay’, we change it to the Rupee symbol. Example in Figure 15 .

If a quantity is found without an entity or container, we retain the same entity and container from the last state and create a new state with the quantity found. Example in Figure 16 .

If the entity in question is not found in states, we assume the entity of the first state to be the entity in question and perform the steps of finding the answer. Example in Figure 17 .

A.2 Examples of Errors made by Solver

Table 4 gives examples of errors made by the rule-based solver.

Error Category	Example
Irrelevant Information	raam is maheene 11 kriket ke maich dekhane gaya. vah pichhale maheene 17 maich dekhane gaya tha aur agale maheene 16 maich dekhane jaaega. vah ab tak kul kitane maich dekh chuka hai?
	Gloss: Ram went to watch 11 cricket matches this month. He went to watch 17 matches last month and next month he will go to watch 16 matches. How many matches has he watched till now?
	Error: Solver returns answer as X=11+17+16. 16 matches that Ram will see next month is irrelevant to the question being asked in the word problem.
Error in entity/container/action identification	shurooaat mein jen ke paas 87 kele the. 7 1 ghode dvaara khae gae. ant mein jen ke paas kitane kele bache?
	Gloss: Initially Jane had 87 bananas. 7 were eaten by 1 horse. How many bananas are left with Jane at the end?
	Error: 7 is not mapped to ‘kele’ by the solver and is therefore missed in calculation.
Set Completion	4 bachchon, 2 karmachaariyon aur 3 adhyaapakon ka 1 samooh chidiyaaghar ja raha hai. chidiyaaghar kitane log ja rahe hain?
	Gloss: 1 group consisting of 4 children, 2 staff and 3 teachers is going to zoo. How many people are going to the zoo?
	Error: Here, bacche (children), karmachaari (staff) and adhyaapak (teachers) form a set - log (people), which solver is not capable of identifying.
Parsing Errors	evalin ke paas shuruaat mein 76 taaaifiyaan theen. kristeen ne evalin ko 72 taaaifiyaan deen. evalin ke paas kitanee taaaifiyaan hain?
	Gloss: Evelyn initially had 76 candies. Christine gave 72 candies to Evelyn. How many candies does Evelyn have?
	Error: ‘taaaifiyaan’ gets tagged as VM i.e. verb in first statement and is missed from calculation.

//-->

	2023-10-12 Word problem Solving is a popular NLP task that deals with solving mathematical problems described in natural language. Mathematical Word Problems cover problems over a large mathematical domain with various complexities ranging from Arithmetic and Algebraic to Geometry and Calculus. While most word problems are entirely textual, some word problems, like geometric word problems, may also have a visual component. Much research has been carried out to solve different genres of word problems with various complex- ity levels in recent years. However, most publicly available datasets and work are in English. Recently there has been a surge in word problem-solving in Chinese with the creation of large benchmark datasets. Labelled benchmark datasets for low-resource languages are very scarce. The first requirement for solving word problems, like any other problem, is data. To the best of our knowledge, no datasets are available for any Indian Languages for Word Problem Solving. Such limitations on data availability not only encouraged us to create a new dataset for an Indian Language but also made us explore techniques by which data for other Indian Languages can be created with ease. In this work, we present a diverse dataset containing 2336 Arithmetic word problems in Hindi built by manually crafting word problems and using word problems augmented from benchmark datasets of other languages. For augmentation, we used translation of word problems (of other languages in which Word Problem Solving data was developed) as a tool to generate diverse word problems. In this process, we studied the translated word problems, gathered the patterns of issues seen in these translations and defined the steps to eliminate the errors and improve the quality of the translated output for it to be suited to be a set of Hindi word problems that look as natural as word problems that are studied across India in Hindi medium schools. We also developed baseline systems for solving these word problems - a rule-based solver that uses verbs to identify operations for generating the answers to word problems and an end-to-end deep learning-based solver that generates equations for word problems. We also propose a new evaluation technique for word problem solvers taking equation equivalence into account. This will form the basis for future work for Word Problem Solving in Indian Languages, especially in Hindi. Centre for Language Technologies Research Centre
Copyright © 2009 - . All Rights Reserved.

Verb Categorisation for Hindi Word Problem Solving

• Sharma, Harshita
• Mishra, Pruthwik
• Misra Sharma, Dipti

Word problem Solving is a challenging NLP task that deals with solving mathematical problems described in natural language. Recently, there has been renewed interest in developing word problem solvers for Indian languages. As part of this paper, we have built a Hindi arithmetic word problem solver which makes use of verbs. Additionally, we have created verb categorization data for Hindi. Verbs are very important for solving word problems with addition/subtraction operations as they help us identify the set of operations required to solve the word problems. We propose a rule-based solver that uses verb categorisation to identify operations in a word problem and generate answers for it. To perform verb categorisation, we explore several approaches and present a comparative study.

• Computer Science - Computation and Language;
• Computer Science - Artificial Intelligence;

Navigation Menu

Search code, repositories, users, issues, pull requests..., provide feedback.

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly.

To see all available qualifiers, see our documentation .

• Notifications You must be signed in to change notification settings

This repository provides HAWP: a dataset for Hindi Word Problem Solving and a baseline (LREC 2022)

hellomasaya/hawp

Folders and files.

Name		Name
4 Commits

Repository files navigation

Hindi arithmetic word problems - hawp.

This repository provides HAWP(a first ever diverse dataset for Hindi) for evaluating and developing MWP Solvers for Hindi. It contains 2336 Hindi Math Word Problems (MWPs), and is published in the paper "HAWP: a Dataset for Hindi Arithmetic Word Problem Solving".

The dataset can be found in hawp/dataset/

About the dataset

Name	HAWP
Version	v1.0
Resource Type	Corpus
Language Type	Monolingual
Language	Hindi
Size	2336 MWPs
Modality	Written
Annotation	Equation, Relevant Indices, Number Of Operators
Use	For Automatic Word Problem Solving
Format	JSON

Snapshot of dataset:

• pIndex : A unique identifier for a MWP
• Problem : The word problem text (includes statements and clues for solving an MWP) and question to be solved based on the word problem text.
• Equation : The equation used to solve the MWP
• Relevant Indices : List of incides of quantities in the MWP that are relevant to solve the question. Whenever the MWP requires world knowledge or uses implict quantities, the relevant indices list contains implicit keyword.
• Number of Operators : Number of operations required to solve the MWP.

• Corpus ID: 252357293

HAWP: a Dataset for Hindi Arithmetic Word Problem Solving

• Harshita Sharma , Pruthwik Mishra , D. Sharma
• Published in International Conference on… 2022
• Computer Science, Mathematics, Linguistics

Figures and Tables from this paper

21 References

Equgener: a reasoning network for word problem solving by generating arithmetic equations, ape210k: a large-scale and template-rich dataset of math word problems, are nlp models really able to solve simple math word problems.

• Highly Influential

How well do Computers Solve Math Word Problems? Large-Scale Dataset Construction and Evaluation

Mathqa: towards interpretable math word problem solving with operation-based formalisms, mawps: a math word problem repository, solving arithmetic word problems automatically using transformer and unambiguous representations, a diverse corpus for evaluating and developing english math word problem solvers, solving general arithmetic word problems, mapping to declarative knowledge for word problem solving, related papers.

Showing 1 through 3 of 0 Related Papers

• Applied research
• Smart City Living Lab
• Technology and Society
• IIITH Newsletter – June 2024
• IIITH Newsletter – May 2024
• IIITH Newsletter – April 2024
• IIITH Newsletter – March 2024
• IIITH Newsletter – February 2024
• IIITH Newsletter – January 2024
• IIITH Newsletter – December 2023
• IIITH Newsletter – November 2023
• IIITH Newsletter – October 2023
• IIITH Newsletter – September 2023
• IIITH Newsletter – August 2023
• IIITH Newsletter – July 2023
• IIITH Newsletter – June 2023
• IIITH Newsletter – May 2023
• IIITH Newsletter – April 2023
• IIITH Newsletter – March2023
• IIITH Newsletter – February 2023
• IIITH Newsletter – January 2023
• IIITH Newsletter – December2022
• IIITH Newsletter – November 2022
• IIITH Newsletter – October 2022
• IIITH Newsletter – September 2022
• IIITH Newsletter – August 2022
• IIITH Newsletter – July 2022
• IIITH Newsletter – June 2022
• IIITH Newsletter – May 2022
• IIITH Newsletter – April 2022
• IIITH Newsletter – March 2022
• IIITH Newsletter – February 2022
• IIITH Newsletter – January 2022
• IIITH Newsletter – December 2021
• IIITH Newsletter – November 2021
• IIITH Newsletter – October 2021
• IIITH Newsletter – September 2021
• IIITH Newsletter – August 2021
• IIITH Newsletter – July 2021
• IIITH Newsletter – June 2021
• IIITH Newsletter – May 2021
• IIITH Newsletter – April 2021
• IIITH Newsletter – March 2021
• IIITH Newsletter – February 2021
• IIITH Newsletter – January 2021
• IIITH Newsletter – December 2020
• IIITH Newsletter – November 2020
• IIITH Newsletter – October 2020
• IIITH Newsletter – September 2020
• IIITH Newsletter – August 2020
• IIITH Newsletter – July 2020
• IIITH Newsletter – June 2020
• IIITH Newsletter – May 2020
• IIITH Newsletter – April 2020
• IIITH Newsletter – March 2020
• IIITH Newsletter – February 2020
• IIITH Newsletter – January 2020
• IIITH Newsletter – December 2019
• IIITH Newsletter – November 2019
• October 2019 Newsletter
• September 2019 Newsletter
• August 2019 Newsletter
• July 2019 Newsletter
• June 2019 Newsletter
• May 2019 Newsletter
• April 2019 Newsletter
• March 2019 Newsletter
• February 2019 Newsletter
• January 2019 Newsletter
• December 2018 Newsletter
• November 2018 Newsletter
• October 2018 Newsletter
• September 2018 Newsletter
• August 2018 Newsletter
• July 2018 Newsletter
• June 2018 Newsletter
• May 2018 Newsletter
• April 2018 Newsletter
• March 2018 Newsletter
• February 2018 Newsletter
• January 2018 Newsletter
• December 2017 Newsletter

Harshita Sharma – Hindi Word Problem

Harshita Sharma, s upervised by Prof. Dipti M Sharma received her Master of Science – Dual Degree in  Computational Linguistics (CL). Here’s a summary of her research work on Hindi Word Problem Solving:

Word problem Solving is a popular NLP task that deals with solving mathematical problems described in natural language. Mathematical Word Problems cover problems over a large mathematical domain with various complexities ranging from Arithmetic and Algebraic to Geometry and Calculus. While most word problems are entirely textual, some word problems, like geometric word problems, may also have a visual component. Much research has been carried out to solve different genres of word problems with various complexity levels in recent years. However, most publicly available datasets and work are in English. Recently there has been a surge in word problem-solving in Chinese with the creation of large benchmark datasets. Labelled benchmark datasets for low-resource languages are very scarce. The first requirement for solving word problems, like any other problem, is data. To the best of our knowledge, no datasets are available for any Indian Languages for Word Problem Solving. Such limitations on data availability not only encouraged us to create a new dataset for an Indian Language but also made us explore techniques by which data for other Indian Languages can be created with ease. In this work, we present a diverse dataset containing 2336 Arithmetic word problems in Hindi built by manually crafting word problems and using word problems augmented from benchmark datasets of other languages. For augmentation, we used translation of word problems (of other languages in which Word Problem Solving data was developed) as a tool to generate diverse word problems. In this process, we studied the translated word problems, gathered the patterns of issues seen in these translations and defined the steps to eliminate the errors and improve the quality of the translated output for it to be suited to be a set of Hindi word problems that look as natural as word problems that are studied across India in Hindi medium schools. We also developed baseline systems for solving these word problems – a rule-based solver that uses verbs to identify operations for generating the answers to word problems and an end-to-end deep learning-based solver that generates equations for word problems. We also propose a new evaluation technique for word problem solvers taking equation equivalence into account. This will form the basis for future work for Word Problem Solving in Indian Languages, especially in Hindi.

October 2023

Subscribe to the PwC Newsletter

Join the community, edit social preview.

Add a new code entry for this paper

Remove a code repository from this paper, mark the official implementation from paper authors, add a new evaluation result row.

TASK	DATASET	MODEL	METRIC NAME	METRIC VALUE	GLOBAL RANK	REMOVE

Remove a task

Add a method, remove a method, edit datasets, hawp: a dataset for hindi arithmetic word problem solving.

LREC 2022  ·  Harshita Sharma , Pruthwik Mishra , Dipti Sharma · Edit social preview

Word Problem Solving remains a challenging and interesting task in NLP. A lot of research has been carried out to solve different genres of word problems with various complexity levels in recent years. However, most of the publicly available datasets and work has been carried out for English. Recently there has been a surge in this area of word problem solving in Chinese with the creation of large benchmark datastes. Apart from these two languages, labeled benchmark datasets for low resource languages are very scarce. This is the first attempt to address this issue for any Indian Language, especially Hindi. In this paper, we present HAWP (Hindi Arithmetic Word Problems), a dataset consisting of 2336 arithmetic word problems in Hindi. We also developed baseline systems for solving these word problems. We also propose a new evaluation technique for word problem solvers taking equation equivalence into account.

Code Edit Add Remove Mark official

Tasks edit add remove, datasets edit, results from the paper edit add remove, methods edit add remove.

Verb Categorisation for Hindi Word Problem Solving

Word problem Solving is a challenging NLP task that deals with solving mathematical problems described in natural language . Recently, there has been renewed interest in developing word problem solvers for Indian languages. As part of this paper, we have built a Hindi arithmetic word problem solver which makes use of verbs. Additionally, we have created verb categorization data for Hindi. Verbs are very important for solving word problems with addition/subtraction operations as they help us identify the set of operations required to solve the word problems. We propose a rule-based solver that uses verb categorisation to identify operations in a word problem and generate answers for it. To perform verb categorisation, we explore several approaches and present a comparative study.

We're not able to analyze this paper right now due to high demand.

Please check back later (sorry!).

Generate a summary of this paper on our Premium plan:

We ran into a problem analyzing this paper.

• Unit Dependency Graph and its Application to Arithmetic Word Problem Solving
• Mapping to Declarative Knowledge for Word Problem Solving
• Arithmetic Word Problem Solver using Frame Identification
• CLEVR-Math: A Dataset for Compositional Language, Visual and Mathematical Reasoning
• Explaining Math Word Problem Solvers

Get summaries of trending comp sci papers delivered straight to your inbox:

Unsubscribe anytime.

If you're seeing this message, it means we're having trouble loading external resources on our website.

If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.

To log in and use all the features of Khan Academy, please enable JavaScript in your browser.

Class 6 Math (India) - Hindi

Course: class 6 math (india) - hindi   >   unit 3, gcf & lcm word problems (hindi).

• GCF & LCM word problems

Want to join the conversation?

Video transcript.

Fact Check: Kamala Harris Did NOT Say 'A Problem Is Not A Problem'

• Aug 2, 2024
• by: Uliana Malashenko

Did Vice President Kamala Harris say "a problem is not a problem" as part of a longer quote in a meme circulating on social media? No, that's not true: A spokesperson for the vice president's office answered "no" by email when Lead Stories asked if the quote was real. Lead Stories could find no video or news articles indicating Harris ever said the words in the meme.

The claim resurfaced in a post (archived here ) on Instagram on July 27, 2024. It showed an image of Harris with the following inscription:

THE PROBLEM OF SOLVING A PROBLEM IS NOT A PROBLEM, BUT WHEN A PROBLEM SOLVES A PROBLEM WITHOUT ANY PROBLEM, THEN THE PROBLEM IS NOT AT ALL A PROBLEM.

When asked whether Harris made the comment that is the focus of this fact check, a spokesman for the vice president's office told Lead Stories via email on August 2, 2024, that the answer is "no."

The quote in question started to circulate on social media on January 20, 2024. Such examples can be found on X, formerly known as Twitter (archived here ), and Reddit (archived here ). Those entries did not contain any elaboration about when and where Harris supposedly made the cited remark.

By January 2024, she had already been in the public eye for decades, including the first three years of serving as vice president.

A Google search (archived here ) across U.S. government websites for the fragment of the purported quote did not produce any results. Had Harris made the comment posted on Instagram and elsewhere, it would likely have been reflected in official transcripts.

A search for the same fragment across news publications indexed by Google News (archived here ) did not lead to any credible reporting confirming the quote. It did show that Snopes has reviewed the same claim.

Lead Stories previously debunked other claims that misattributed similar quotes to Harris. For example, she did not say that "Olympians are Olympic ."

Other Lead Stories fact checks mentioning Kamala Harris can be found here . Stories about the 2024 U.S. presidential election are here .

Uliana Malashenko is a  New York-based freelance writer and fact checker. Read more about or contact Uliana Malashenko

Lead Stories is a fact checking website that is always looking for the latest false, misleading, deceptive or inaccurate stories, videos or images going viral on the internet. Spotted something? Let us know! .

Lead Stories is a:

• Verified signatory of the IFCN Code of Principles
• Facebook Third-Party Fact-Checking Partner
• Member of the #CoronavirusFacts Alliance

@leadstories

Subscribe to our newsletter

Please select all the ways you would like to hear from Lead Stories LLC:

You can unsubscribe at any time by clicking the link in the footer of our emails. For information about our privacy practices, please visit our website.

We use Mailchimp as our marketing platform. By clicking below to subscribe, you acknowledge that your information will be transferred to Mailchimp for processing. Learn more about Mailchimp's privacy practices here.

Fact Check: FAKE Image Of Hitler Used In Meme To Compare Trump With Hitler In Raised-Fist Composite

Fact Check: Poll Managers In Myrtle Beach Have NO Reason To Write On Ballot And It Will NOT Be Disqualified If A Poll Manager Writes On It

Fact Check: Kamala Harris Descended From A Slave Owner, But It's Not 'An Inconvenient Part Of Her History' Or A 'Deep, Dark Secret' That 'Could Come Back To Haunt Her'

Fact Check: Video Does NOT Show Olympic Athlete Vomiting Because He Swam In Seine -- He Cited Heat

Fact Check: NO Evidence Female Algerian Boxer Imane Khelif Used To Be Man

Fact Check: Elon Musk Did NOT Announce That He Would 'Stop All Donations' To Olympics -- Fake Quote Is From AI-Produced Article

Fact Check: Photo Does NOT Show Undamaged Ear In Immediate Weeks After Trump Shooting

Most recent.

Fact Check: Man in Meme Did NOT Rape 12-Year-Old But Was Part of Gang That Did -- He Was Convicted for Sexual Assault

Fact Check: Turkish Olympian Yusuf Dikec Did NOT Spontaneously Get Interested In Sport Shooting During Divorce

Fact Check: Mike Tyson Did NOT Say He's Willing To Box 'Olympic Dude'

Fact Check: Jack Black Did NOT Receive Lifetime Ban From Academy Of Country Music Following Bandmate's Comments About Trump

Fact Check: Project 2025 Does NOT Claim 'Only Valid Family' Is Working Father, Stay-At-Home Mother, Their Children

Fact Check: NO 'Leaked Footage' Of Trump Shooter And BlackRock CEO Holding Gun -- FAKE AI-Generated Still Images

Fact Check: NO Evidence Olympic Gold Medalist Simone Biles Is Retiring Or 'Expecting A Baby' As 2024 Paris Games Wrap Up

Share your opinion, lead stories.

Sign in to add this item to your wishlist, follow it, or mark it as ignored

	Interface	Full Audio	Subtitles
English	✔
Japanese	✔

Buy Reimu Needs Help!? Aunn-chan to the Rescue!

INTRODUCTORY OFFER! Offer ends in

About This Game

System requirements.

• OS: Windows10/11

More like this

What curators say, customer reviews.

You can use this widget-maker to generate a bit of HTML that can be embedded in your website to easily allow customers to purchase this game on Steam.

Enter up to 375 characters to add a description to your widget:

Copy and paste the HTML below into your website to make the above widget appear

Popular user-defined tags for this product: (?)

Sign in to add your own tags to this product.