Language is one of the most important landmarks in humans in history. However, most languages could be ambiguous, which means the same conveyed text or speech, results in different actions by different readers or listeners. In this project we propose a method to detect the ambiguity of a sentence using translation by multilingual LLMs. In this context, we hypothesize that a good machine translator should preserve the ambiguity of sentences in all target languages. Therefore, we investigate whether ambiguity is encoded in the hidden representation of a translation model or, instead, if only a single meaning is encoded. The potential applications of the proposed approach span i) detecting ambiguous sentences, ii) fine-tuning existing multilingual LLMs to preserve ambiguous information, and iii) developing AI systems that can generate ambiguity-free languages when needed.
Report
Share
Report
Share
1 of 26
Download to read offline
More Related Content
Similar to Detecting and translating language ambiguity with multilingual LLMs
The document discusses learning lines, which refer to the construction of knowledge and skills throughout the curriculum from basic to more complex. It provides examples of how geospatial thinking and skills can be developed as a continuum from easy to advanced abilities, such as interpreting maps to capturing and analyzing spatial data. The concept of "local" is also examined from different perspectives like varying definitions and how places are interconnected rather than isolated.
Commonsense knowledge for Machine Intelligence - part 1Niket Tandon
This document summarizes a research tutorial on commonsense knowledge for machine intelligence that will take place from November 6-10, 2017 in Singapore. The tutorial covers three main parts: acquiring commonsense knowledge, detecting and correcting odd collocations in text, and applications and open issues. It presents various commonsense knowledge bases that have been developed using both manual and automated methods to represent different types of commonsense knowledge in both discrete and continuous representations.
Big Data Palooza Talk: Aspects of Semantic ProcessingNa'im Tyson
This inaugural, Meetup talk, sponsored by the Knowledgent Group, discussed aspects of semantic processing, and emphasized using python for lexical semantics. Slides cite example code snippets for computing the relationships between words using the Natural Language Toolkit (NLTK) in Python. There is also a small overview of the technologies underlying the Semantic Web and text mining.
Text Processing Like Humans Do: Visually Attacking and Shielding NLP SystemsSatoshi Akasaki
This paper develops three methods for visually perturbing text to attack NLP systems, confirms that humans are robust to such perturbations but NLP performance degrades, and proposes three methods for shielding systems from visual attacks. Specifically:
- It introduces three visual perturbation methods: image-based, description-based, and easy-character replacements.
- Human tests show humans have low error rates understanding perturbed text, while NLP tasks like POS tagging and toxicity classification suffer reduced accuracy.
- To shield systems, it employs adversarial training, initializing models with visual embeddings, and rule-based character recovery - which improve performance over unshielded models under attack.
Deep generative and discriminative models for speech recognition. The document outlines the history of speech recognition models including early neural networks, hidden dynamic models, and deep belief networks. It describes how deep learning entered speech recognition around 2009 through the collaboration of Microsoft Research and academics. This led to replacing generative models with discriminative deep neural networks which achieved large error reductions. The talk outlines further innovations in deep learning for speech including context-dependent models and better optimization techniques.
The document discusses image captioning using deep neural networks. It begins by providing examples of how humans can easily describe images but generating image captions with a computer program was previously very difficult. Recent advances in deep learning, specifically using convolutional neural networks (CNNs) to recognize objects in images and recurrent neural networks (RNNs) to generate captions, have enabled automated image captioning. The document discusses CNN and RNN architectures for image captioning and provides examples of pre-trained models that can be used, such as VGG-16.
Building a Neural Machine Translation System From ScratchNatasha Latysheva
Human languages are complex, diverse and riddled with exceptions – translating between different languages is therefore a highly challenging technical problem. Deep learning approaches have proved powerful in modelling the intricacies of language, and have surpassed all statistics-based methods for automated translation. This session begins with an introduction to the problem of machine translation and discusses the two dominant neural architectures for solving it – recurrent neural networks and transformers. A practical overview of the workflow involved in training, optimising and adapting a competitive neural machine translation system is provided. Attendees will gain an understanding of the internal workings and capabilities of state-of-the-art systems for automatic translation, as well as an appreciation of the key challenges and open problems in the field.
This document describes an SVM approach called MiniExperts for measuring semantic textual similarity that was submitted to SemEval 2015 Task 2. It extracts various features from text such as part-of-speech tags, lemmatization, named entities, and multiword expressions. It also calculates conceptual, semantic, and distributional similarity measures between sentences. When applied to the SemEval task, it achieved a Pearson correlation of 0.7216 for English and 0.5158 for Spanish, ranking 33rd and 9th respectively out of the submitted systems. Future work will focus on extracting conceptual descriptions from BabelNet to further improve matching of conceptual terms between sentences.
Sequential Reptile_Inter-Task Gradient Alignment for Multilingual LearningMLAI2
Multilingual models jointly pretrained on multiple languages have achieved remarkable performance on various multilingual downstream tasks. Moreover, models finetuned on a single monolingual downstream task have shown to generalize to unseen languages. In this paper, we first show that it is crucial for those tasks to align gradients between them in order to maximize knowledge transfer while minimizing negative transfer. Despite its importance, the existing methods for gradient alignment either have a completely different purpose, ignore inter-task alignment, or aim to solve continual learning problems in rather inefficient ways. As a result of the misaligned gradients between tasks, the model suffers from severe negative transfer in the form of catastrophic forgetting of the knowledge acquired from the pretraining. To overcome the limitations, we propose a simple yet effective method that can efficiently align gradients between tasks. Specifically, we perform each inner-optimization by sequentially sampling batches from all the tasks, followed by a Reptile outer update. Thanks to the gradients aligned between tasks by our method, the model becomes less vulnerable to negative transfer and catastrophic forgetting. We extensively validate our method on various multi-task learning and zero-shot cross-lingual transfer tasks, where our method largely outperforms all the relevant baselines we consider.
This document summarizes a paper on using simple lexical overlap features with support vector machines (SVMs) for Russian paraphrase identification. It introduces paraphrase identification and various paraphrase corpora. It then describes a knowledge-lean approach using only tokenization, lowercasing, and overlap features like union and intersection size as inputs to linear and RBF kernel SVMs. The method achieves competitive results on English, Turkish, and Russian paraphrase identification tasks.
Machine Learning Foundations for Professional ManagersAlbert Y. C. Chen
20180526@Taiwan AI Academy, Professional Managers Class.
Covering important concepts of classical machine learning, in preparation for deep learning topics to follow. Topics include regression (linear, polynomial, gaussian and sigmoid basis functions), dimension reduction (PCA, LDA, ISOMAP), clustering (K-means, GMM, Mean-Shift, DBSCAN, Spectral Clustering), classification (Naive Bayes, Logistic Regression, SVM, kNN, Decision Tree, Classifier Ensembles, Bagging, Boosting, Adaboost) and Semi-Supervised learning techniques. Emphasis on sampling, probability, curse of dimensionality, decision theory and classifier generalizability.
GPT-2: Language Models are Unsupervised Multitask LearnersYoung Seok Kim
This document summarizes a technical paper about GPT-2, an unsupervised language model created by OpenAI. GPT-2 is a transformer-based model trained on a large corpus of internet text using byte-pair encoding. The paper describes experiments showing GPT-2 can perform various NLP tasks like summarization, translation, and question answering with limited or no supervision, though performance is still below supervised models. It concludes that unsupervised task learning is a promising area for further research.
This paper aims to develop an effective sentence model using a dynamic convolutional neural network (DCNN) architecture. The DCNN applies 1D convolutions and dynamic k-max pooling to capture syntactic and semantic information from sentences with varying lengths. This allows the model to relate phrases far apart in the input sentence and draw together important features. Experiments show the DCNN approach achieves strong performance on tasks like sentiment analysis of movie reviews and question type classification.
This document describes research applying techniques from program analysis to automatically infer properties of code in the Maple computer algebra system. The researchers developed abstract interpretation frameworks tailored to Maple and used these to gather constraints about Maple code and values. By analyzing the entire Maple library with this approach, they were able to infer simple but useful properties, like type information, for parts of the library. This demonstrated the potential of applying formal methods to understand and reason about large, dynamically-typed code bases.
This document presents research on developing an automatic speech recognition system for Egyptian Arabic. It explores using a diacritized lexicon versus an undiacritized graphemic lexicon, and compares different acoustic models. The researchers used the Al-Jazeera Egyptian dialect corpus and evaluated performance using word error rate. Their results showed that a graphemic lexicon outperformed a diacritized lexicon, and that reducing out-of-vocabulary words and using deep neural network acoustic models improved word error rate.
Data Generation with PROSPECT: a Probability Specification ToolIvan Ruchkin
The document describes PROSPECT, a tool for automatically generating synthetic data from probabilistic specifications. PROSPECT allows users to specify discrete temporal distributions declaratively and solves the specifications algebraically to obtain unique parameter solutions. If a unique distribution is defined, PROSPECT can sample from it. The document outlines PROSPECT's specification language and approach, which involves parameterizing specifications, solving systems of equations, and sampling solutions. An evaluation found PROSPECT specifications were more succinct than probabilistic programs and generated accurate data.
Similar to Detecting and translating language ambiguity with multilingual LLMs (20)
Dalghren, Thorne and Stebbins System of Classification of AngiospermsGurjant Singh
The Dahlgren, Thorne, and Stebbins system of classification is a modern method for categorizing angiosperms (flowering plants) based on phylogenetic relationships. Developed by botanists Rolf Dahlgren, Robert Thorne, and G. Ledyard Stebbins, this system emphasizes evolutionary relationships and incorporates extensive morphological and molecular data. It aims to provide a more accurate reflection of the genetic and evolutionary connections among angiosperm families and orders, facilitating a better understanding of plant diversity and evolution. This classification system is a valuable tool for botanists, researchers, and horticulturists in studying and organizing the vast diversity of flowering plants.
ALTERNATIVE ANIMAL TOXICITY STUDY .pptxSAMIR PANDA
Alternatives animal testing are development and implementation of test methods that avoid the use of live animals.
Human biochemistry, physiology, pharmacology, and endocrinology and toxicology has been derived from animal models.10-100 millions of animals are using for experimentation in a year.
Animals used experimentation distributed among zebra- fish to primates.
Vast majority of animals are sacrificed at end of research programme.The use of animals can be further subdivided according to the degree of suffering
Minor animal suffering:- observing animals in behavioral studies, single blood sampling, Immunization without adjutants, etc.
Moderate animal suffering:- repeated blood sampling, recovery from general anesthesia, etc.
Search for Dark Matter Ionization on the Night Side of Jupiter with CassiniSérgio Sacani
We present a new search for dark matter (DM) using planetary atmospheres. We point out that
annihilating DM in planets can produce ionizing radiation, which can lead to excess production of
ionospheric Hþ
3 . We apply this search strategy to the night side of Jupiter near the equator. The night side
has zero solar irradiation, and low latitudes are sufficiently far from ionizing auroras, leading to a lowbackground search. We use Cassini data on ionospheric Hþ
3 emission collected three hours either side of
Jovian midnight, during its flyby in 2000, and set novel constraints on the DM-nucleon scattering cross
section down to about 10−38 cm2. We also highlight that DM atmospheric ionization may be detected in
Jovian exoplanets using future high-precision measurements of planetary spectra.
Science-9-Lesson-1 ang lesson 2-NLC-pptx.pptxJoanaBanasen1
just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it! just check it!
CULEX MOSQUITOES, SYSTEMATIC CLASSIFICATION, MORPHOLOGY, LIFE CYCLE , CLINICA...DhakeshworShougrakpa
showing Culex mosquitoes' systematic classification, a completed life cycle i.e. egg, larva, pupa and adult mosquitoes also known as imago, also this slide showed the morphology of culex mosquitoes including head, thorax, abdomen, wing, egg larval stage, resting position,etc. by comparing with anopheles' mosquitoes. it's also showed the transmission of wuchereria bancrofti transmitted by vector Culex quinquefasciatus. Host: W. bancrofti completes its life cycle in
two hosts.
1. Definitive host: Man
2. Intermediate host: Mosquito named
Culex quinquefasciatus is the principle
vector worldwide. Rarely Anopheles
(rural Africa) or Aedes (Pacific Island)
can serve as a vector.
Infective form: Third stage filariform larvae
are the infective form found in the proboscis
of the mosquito.
Mode of transmission: L3
filariform larvae get
deposited in skin by the insect bite. Residents living in the endemic areas are exposed to
about 50–300 L3
larvae every year.Human cycle
z Develop into adults: Larvae penetrate
the skin, enter into lymphatic vessels and
migrate to the local lymph nodes where they
molt twice to develop into adult worms in
few months (4–6 weeks for B. malayi)
z Adults lay L1
larvae (microfilariae): Adult
worms reside in the afferent lymphatics or
cortical sinuses of the lymph nodes where
they mate and start laying the first stage
larvae (microfilariae). Male worms die after
mating where as the female worms live for
5–10 years. A gravid female can discharge
50,000 microfilariae/day
z Prepatent period: It is the time period
between the infection (entry of L3
larvae)
and diagnosis (detection of microfilariae
in blood). This is variable ranging from 80
days to 150 days
Mosquito cycle
z Transmission: When the mosquito bites
an infected man, the microfilariae are
ingested. Culex bites in night where as Aedes
bites in daytime
z Exsheathing: Microfilariae come out of the
sheath within 1–2 hours of ingestion
z Migration to thoracic muscle: L1
larvae
penetrate the stomach wall and migrate to
thoracic muscle in 6–12 hours where they
become sausage shaped (short and thick)
z Develop to infective L3
larvae: L1
larvae
molt twice to develop L2
(long and thick
form) followed by L3
(long and thin form).
The highly active L3
larvae migrate to the
labella (distal part of proboscis) of the
mosquito and serve as the infective stage
to man
z Extrinsic incubation period: Under
optimum conditions, the mosquito cycle
takes around 10–14 days
Clinical symptoms:
The clinical symptoms and signs are mainly determined by the duration of the infection. The
adult worms, which live in the lymphatic vessels, can cause severe inflammation of the
lymphatic system and acute recurrent fever. Secondary bacterial infections are a major factor in
the progression towards lymphoedema and elephantiasis, the characteristic swelling of the limbs,
genitalia and breasts.
treatment like using larvicide like fenthion can spray on water
Types of Garden (Mughal and Buddhist style)saloniswain225
Garden is the place where, flower blooming on a plant ,aesthetic things are present like Topiary, Hedges, Arches and many more. Whereas, Botanical garden is an educational institution for scientific research as well as gathering information about different culture. Such as, Hindu, Mughal , Buddhist style.
History & overview of Bioprocess Technology.pptxberciyalgolda1
Bioprocess technology is a field that merges biology, chemistry, and engineering to develop processes that harness living cells or their components (like enzymes) for the production of pharmaceuticals, chemicals, food, and biofuels. This multidisciplinary field has evolved significantly over the past few decades, playing a crucial role in various industries.
Cause and solution of Water hyacinth (Terror of Bengal)saloniswain225
Water hyacinth is a buoyant plant that has a clump of leaves with squishy stalks arising from the lower feathery roots and it is an invasive species came from South America
Prototype Implementation of Non-Volatile Memory Support for RISC-V Keystone E...LenaYu2
Handling confidential information has become an increasingly important concern among many areas of society. However, current computing environments have been still vulnerable to various threats, and we should think they are untrusted.
Trusted Execution Environments (TEEs) have attracted attention because they can execute a program in a trusted environment constructed on an untrusted platform.
Particularly, the RISC-V Keystone is one of the interesting TEEs since it is a flexibly customizable and fully open-source platform. On the other hand, as same as other TEEs, it must also delegate I/O processing, such as file accesses, to a host OS, resulting in the expensive overhead. For this problem, we thought utilizing byte-addressable non-volatile memory (NVM) modules is a useful solution to handle persistent data objects for TEEs.
In this paper, we introduce a prototype implementation of NVM support for the Keystone. Additionally, we evaluate it on the Freedom U500 built on a VC707 FPGA dev kit.
https://ken.ieice.org/ken/paper/20210720TC4K/
PART 1 The New Natural Principles of Electromagnetism and Electromagnetic Fie...Thane Heins
Document Summary and the History of Perpetual Motion
Every single Faraday Generator coil since 1834 has been and is currently performing Negative Work at infinite efficiency with created Electromagnetic Field Energy during electricity generation and its physical Kinetic Energy reduction or Electromagnetic Resistance of the changing magnetic field which is initially inducing Electric Current in the generator coil according to Faraday's Law of Induction.
The Work-Energy Principle confirms mathematically that the magnitude of the changing magnetic field's Kinetic Energy reduction is equal to the magnitude of Negative Work performed at infinite efficiency, which is equal to the magnitude of Energy (Electromagnetic Field Energy which is created according to Oersted's Law of Creation of Energy of 1820). Created Electromagnetic Field Energy is required in order to perform the Negative Work – because Work cannot be performed in the absence of Energy.
In 2007 Thane Heins of Almonte Ontario, Canada discovered that unlimited amounts of Positive Electromechanical Work could be performed at infinite efficiency with created and TIME DELAYED Electromagnetic Field Energy.
Every single ReGenX Generator coil since 2007 has been and is currently performing Positive Work at infinite efficiency with created Electromagnetic Field Energy during electricity generation and during its physical Kinetic Energy increase or Electromagnetic Assistance of the changing magnetic field which is initially inducing Electric Current in the generator coil according to Heins' Law of Induction.
Faraday Electric Generators all harness internally Created Electromagnetic Field Energy in order to perform Negative Work (system Kinetic Energy reduction) at infinite efficiency and ReGenX Electric Generators harness internally created and Time Delayed Electromagnetic Field Energy in order to perform Positive Work (system Kinetic Energy increase) at infinite efficiency.
Both Faraday Generators and ReGenX Generators operate as Perpetual Motion Machines of the First Kind because they both have the ability to perform both Negative or Positive Work indefinitely and at infinite efficiency without requiring any External Energy input. The unlimited Energy required to perform either the Negative or Positive Work is created at the Sub-Atomic Quantum Electron level inside the generators' Current Bearing Wires according to the Law of Creation of Energy.
Hans Christian Oersted discovered the Law of Creation of Energy in 1820 when he demonstrated the world's first Perpetual Motion Machine of the First Kind at the University of Copenhagen when he also simultaneously violated Newton's 1st, 2nd and 3rd Laws of Motion.
Michael Faraday built and demonstrated the world's second Perpetual Motion Machine of the First Kind in 1822 when he demonstrated his Electric Motor invention which harnessed created Electromagnetic Field Energy in order to perform Positive Electromechanical Work at infinite efficienc
Principles of Colorimetry - Mastering the Art of Colour MeasurementColours Guide
Unlock the secrets of colour science with our comprehensive presentation on the principles of colorimetry. This guide delves into the critical aspects of colour measurement, essential for ensuring consistent quality and customer satisfaction across various industries.
Key Topics Covered:
- Introduction to Colorimetry
- Human Perception of Colour
- Importance of Metamerism and Colour Constancy
- Understanding Different Colour Spaces (RGB, CMYK, Lab)
Utilising Spectrophotometers and Colorimeters
- Implementing Quality Control for Colour Consistency
Learn how mastering these techniques can enhance product quality and meet customer expectations. Perfect for professionals in printing, textiles, paint, food, cosmetics, automotive, electronics, and more.
Read the full article: https://colours.guide/principles-of-colorimetry/
Visit www.colours.guide for more insights and resources on colour science.
3. WHAT IS AMBIGUITY?
3
Mariano Ceccato, Nadzeya Kiyavitskaya, Nicola Zeni, Luisa Mich, and Daniel M Berry. 2004. Ambiguity identification and
measurement in natural language texts. Publisher: University of Trento.
4. EXAMPLES
• “Give me the bat!” ~ Lexical
• “The professor said on Monday he would give an exam” ~ Syntactic
• “Jane saw the man with a telescope” ~ Semantic
• “I like you too!” ~ Pragmatic
• “The prof said she would give us all A’s.” ~ Vagueness
• “Proposal” to “voorstel” and “aanzoek” ~ Translational
4
Apurwa Yadav, Aarshil Patel, and Manan Shah. 2021b. A comprehensive review on resolving ambiguities in natural language
processing. AI Open, 2:85–92.
5. RESEARCH QUESTIONS
Question 1:
Can a state-of-the-art Transformer-based MT model properly encode whether a sentence in
the source language is (non-)ambiguous?
Question 2:
Are both semantic validity and ambiguity preserved by the translation when the sentence
is translated into a target language, and then translated back?
Question 3:
Can we predict the ambiguity of a sentence by translating it into another language looking
at the learned hidden representations?
5
9. STEP3: MAPPING FUNCTION
Idea:
“Can we map source translation representation to
target translation representation?”
“What are the properties of this mapping function?”
9
10. STEP3: MAPPING FUNCTION (CONT.)
Idea: Autoencoder as a mapping function!
• Normalized reconstruction error
• Model complexity
10
11. STEP3: MAPPING FUNCTION (CONT.)
Idea:
“An Auto-encoder should behave differently in
mapping ambiguous and unambiguous sentences”
“How reconstruction error changes over model size
and target language is predictive of sentence
ambiguity”
11
12. SAMPLE MAPPINGS
Andrei picked up the chair or the bag and
the telescope (ambiguous)
Andrei picked up the chair, or both the
bag and the telescope (unambiguous)
12
15. STEP 4: CLASSIFICATION
• Input:
1. Single language – All models
2. All languages - Best model
3. Only along languages
4. Whole mapping function info.
• Input variable:
1. Reconstruction error
2. Reconstruction error differences
• Output:
1. Ambiguous vs unambiguous
2. Ambiguity type (4 classes)
• Model
1. Logistic regression
2. Neural network
• Cross-validation
• 10-fold
15
16. RESULTS
Input Input variable Output Model Accuracy F-measure
Engineering
appr.
Alpha Amb. Vs
unamb.
Logistic 52.53 % 0.525
Persian Differences Amb. Vs
unamb.
Logistic 57.81 % 0.578
Best AE Values Amb. Vs
unamb.
Logistic 66.67 % 0.667
Along
languages
Differences Amb. Vs
unamb.
Logistic 85.87 % 0.859
Whole Differences Amb. Type Logistic 92.83 % 0.928
Whole Differences Amb. Vs
unamb.
Logistic 88.19 % 0.882
Best AE Values Amb. Vs
unamb.
NN 73.20 % 0.732
Whole Values Amb. Vs
unamb.
NN 81.99 % 0.820
Whole Values Amb. Type NN 78.26 % -
16
17. FINDINGS
1. Single language translation is not informative enough in predicting ambiguity.
• 57.8059 % 85.865 %
• Adding more features …
2. Single best auto-encoder is not informative enough in predicting ambiguity.
• 66.6667 % 88.1857 %
• Adding more features about the gradual change over the size of the AE model …
3. Adding reconstruction error differences between languages improves accuracy.
• 85.865 % 88.1857 %
• Adding more features about the properties of the mapping function mesh …
17
18. FINDINGS (CONT.)
4. Reconstruction error differences is more informative than their values.
• 81.9876 % 94.9367 %
• The shape of the mapping function is informative not the position …
5. A simple linear model can perform relatively close to a complex NN model.
• 88.1857 % 94.9367 %
• More complex features learned …
6. Predicting more detailed classes improves the accuracy in linear models.
• 88.1857 % 92.827 % (LR)
• More detailed regions in the misclassified regions (see next slide) …
• 94.9367 % 93.038 % (NN)
• Nonlinear boundaries anyway …
18
21. MISCLASSIFICATION EXAMPLES
• Italian (translation problem):
• Source sentence: Andrei and Danny put down the yellow bag and chair (amb.)
• Machine translation: Andrei e Danny mettono il sacchetto giallo e la sedia (unamb.)
• Machine translation back: Andrei and Danny put the yellow bag and the chair. (unamb.)
• Human translation: Andrei e Danny mettono il sacchetto e la sedia color giallo (amb.)
• Persian (language problem):
• Source sentence: Andrei left the person with a green bag (amb.)
• Machine translation: مرد آندری
را
کرد ترک سبز چمدان با (unamb.)
• Machine translation back: Andrew left the man with a green bag. (amb.)
21
22. RESEARCH QUESTIONS
• Question 3: Predicting
• Yes, we can predict language ambiguity using translation with LLMs with high accuracy.
• Question 2: Preservation
• Depends on the language. The problem could be either because of the translation or the
target language itself.
• Question 1: Encoding
• Yes, as ambiguity could be preserved in translation and be predicted as well, the
information shouldn’t have been lost in encoding.
22
23. CONTRIBUTIONS
• Predicting ambiguity without direct use of semantics
• Data complexity: Could be trained with small training data
• Generalizability: Could be easily generalizable to new unseen samples
• Robustness: No need to update classifier model due to change in input distribution
23
24. FUTURE DIRECTIONS
• Interpretability: Detecting source of ambiguity (which word?)
• Extendibility: Extending to source languages other than English
• Analysis: Source of misclassification in all languages via data annotation
24
25. POTENTIAL APPLICATIONS
1. Automatic translation of critical documents e.g. legal, political, commercial, etc.
• Ask the user for an unambiguous sentence
2. Fine-tuning existing multilingual LLMs
• Prevent ambiguity
3. Developing AI systems for generating ambiguity-free languages
(www.Synaptosearch.com)
1. Classify the sentence
2. Get the gradient w.r.t. input
3. Use as a loss function term
25