CALL FOR PAPERS

ACM Transactions on Natural Language Processing for Cross-Modal Learning

Guest Editors:
• Dr. Shadi Mahmoud Faleh AlZu’bi, Al-Zaytoonah University of Jordan, [email protected]
• Dr. Maysam Abbod, Brunel University London, [email protected]
• Dr. Ashraf Darwish, Helwan University, [email protected]

In today’s world, most unstructured data is in the form of text, and structured data is primarily images. Online
websites have millions of product descriptions, mainly in text form, whereas shopping sites already have millions of
pictures. Since it is cumbersome to train the model with textual data and annotate all the images manually, it is
required to learn how to use both available datasets by aligning these two modalities. Such as cross-modal retrieval:
Given an image as input, predict similar-looking books or predict all the books which contain the words written on
the book cover, or indicating book genre given a title of a book along with its summary. For such tasks, one needs
to learn how to build models which can do multimodal learning or cross-modal learning. Natural Language
Processing (NLP) research aims to build systems that understand natural, unconstrained languages in terms of a
finite set of formalized representations. While we can use a search engine to look up information online, it's not
always as good as talking to people and knowing whether they can help or where they're going next. NLP, on the
other hand, NLP is something that we can do very well with our computers. Many people associate it with speech
recognition and the creation of speech-to-text conversion programs for web searches and Wikipedia articles, but it
goes much deeper than that.
An important reason for NLP development is its importance in the realm of cross-modal deep learning. The
multimodal learning model combines data from multiple sources to get better performance than single modal deep
learning. NLP plays a vital role in this context because almost all the information we receive is through natural
language and hence can be used in datasets that utilize multimodal learning models. It is slated to be one of the
most essential and powerful tools for making sense of our data, enabling us to discover new patterns from existing
information and make connections where they weren't apparent before. It supports cross-modal machine learning
where previously, a single application could not analyze diverse information from the data captured from different
sources. Cross-modal Learning helps NLP organizations use natural language processing in understanding inter-
modal data that computer scientists in isolation previously studied. It enables the extraction of relevant information
that can be used to tailor computer communications and can also be applied to cross-modal learning. Though the
primary focus has been on computational linguistics, it has also been extensively studied in cognitive psychology,
computer vision, and semantic computing. NLP will enable advanced learning methods to capture, process, and
analyze all the data needed for training.
The use of NLP for cross-modal learning is particularly relevant as it allows us to train machines with combined input
from multiple modalities. As such, we can incorporate information from sensory systems into more modular
algorithms and curtail over-generalizing errors that often occur due to a lack of training on multiple modalities.
Cross-modal training then makes it possible for specific features in auditory/verbal features like accent etc., visual
elements like lighting and color etc., space surrounding objects, and even tactile features such as surface texture,
via heart-rate variability and respiration synchronized with speech time are all incorporated into the same voice
recognition algorithm which can improve overall performance dramatically. NLP is a growing area of research
aiming to teach computers how to understand the meaning of language. One step in this research direction is
learning to label and make predictions on new data according to existing labeled examples. This requires that tags
be associated with both the data they are to be applied too, as well as an overriding hypothesis of the data's
essence. In order for such a machine learning technique to work, references to important details of the data state
in one modality must also be recognized by processing in other modalities such as images or audio files. This special
issue on Natural Language Processing for cross-modal Learning aims at bringing together researchers from different
communities to exchange ideas, results, and standard models.
Topics

• State-of-the-art in Cross-Modal Learning, focusing on Neural and Statistical methods like Conditional
Random Fields.
• Practical Difficulties and Refinements of Cross-Modal Learning based on recent work in Natural Language
Processing
• Image Annotation and Automatic Captioning
• Learning from Heterogeneous Modalities
• Feature Extraction and Representation in Natural Language Processing
• Multimodal Representation Learning
• Design of Algorithms to exploit both Vision and Language in providing solutions to a Broad Range of Tasks
• Data Descriptives and Transformation for Text Mining
• Coreference Resolution, Sentiment Analysis, and Opinion Mining
• Textual Entailment and Recognizing Textual Entailment (RTE) problems
• Multimodal Information Extraction and Question Answering
• Vector Representation Methods and Evaluation Techniques

Important Dates
• Submissions deadline: May 30, 2024
• First-round review decisions: August 25, 2024
• Deadline for revision submissions: November 30, 2024
• Notification of final decisions: January 25, 2025
• Tentative publication: As per Journal Policy

Submission Information
Please refer https://dl.acm.org/journal/tallip/author-guidelines and select “Natural Language Processing for
Cross-Modal Learning” in the TALLIP submission site, https://mc.manuscriptcentral.com/tallip

For questions and further information, please contact Dr. Shadi Mahmoud Faleh AlZu’bi
/[email protected]