Update projects.md, updates V2

_pages/projects.md

@@ -156,22 +156,20 @@ MSc/BSc thesis research vectors:
 
 <!-- - *Cross-domain Relation extraction.* Extracting structured information such as bornIn(person, location) is central for knowledge base population. Yet current extraction technology is limited to a few text domains and semantic relations. How can we adapt relation extractors to generalize better across different text domains or different relation sets? [See references of MultiVaLUe project](#multivalue). Level: BSc or MSc.--> 
 
-- :hourglass_flowing_sand: *Computational Job Market Analysis.* Job postings are a rich resource to understand the dynamics of the labor market including which skills are demanded, which is also important for an educational viewpoint. Recently, the emerging line of work on computational job market analysis or NLP for human resources has started to provide data resources and models for automatic job posting analysis, such as the identification and extraction of skills. See references of MultiSkill project. For students interested in real-world applications, this theme provides multiple thesis projects including but not limited to: an in-depth analysis of existing data set and models, researching implicit skill extraction or cross-domain transfer learning to adapt skill and knowledge extraction to data sources other than job postings like patents or scientific articles. [See references of MultiSkill project](#multiskill). See also [Bhola et al., 2020](https://aclanthology.org/2020.coling-main.513.pdf) and [Gnehm et al. 2021](https://www.zora.uzh.ch/id/eprint/230653/1/2022.nlpcss_1.2.pdf) and our [own ESCOXLM-R model](https://aclanthology.org/2023.acl-long.662.pdf). Level: BSc or MSc.
+- *Computational Job Market Analysis.* Job postings are a rich resource to understand the dynamics of the labor market including which skills are demanded, which is also important for an educational viewpoint. Recently, the emerging line of work on computational job market analysis or NLP for human resources has started to provide data resources and models for automatic job posting analysis, such as the identification and extraction of skills. See references of MultiSkill project. For students interested in real-world applications, this theme provides multiple thesis projects including but not limited to: an in-depth analysis of existing data set and models, researching implicit skill extraction or cross-domain transfer learning to adapt skill and knowledge extraction to data sources other than job postings like patents or scientific articles. [See references of MultiSkill project](#multiskill). See also [Bhola et al., 2020](https://aclanthology.org/2020.coling-main.513.pdf) and [Gnehm et al. 2021](https://www.zora.uzh.ch/id/eprint/230653/1/2022.nlpcss_1.2.pdf) and our [own ESCOXLM-R model](https://aclanthology.org/2023.acl-long.662.pdf). Level: BSc or MSc.
 
-- :hourglass_flowing_sand: *Climate Change Insights through NLP*. Climate change is a pressing issue internationally that is receiving more and more attention everyday. It is influencing regulations and decision-making in various parts of society such as politics, agriculture, business, and it is discussed extensively on social media. For students interested in real-world societal applications, this project aims to contribute insights on the discussion surrounding climate change on social media by examining discourse from a social media platform. The data will have to be collected (potentially from existing sources), cleaned, and analyzed using NLP techniques to examine various aspects or features of interest such as stance, sentiment, the extraction of key players, etc. References: [Luo et al., 2020](https://aclanthology.org/2020.findings-emnlp.296v2.pdf), [Stede & Patz, 2021](https://aclanthology.org/2021.nlp4posimpact-1.2.pdf), [Vaid et al., 2022](https://aclanthology.org/2022.acl-srw.35.pdf). Level: BSc or MSc.
+- *Climate Change Insights through NLP*. Climate change is a pressing issue internationally that is receiving more and more attention everyday. It is influencing regulations and decision-making in various parts of society such as politics, agriculture, business, and it is discussed extensively on social media. For students interested in real-world societal applications, this project aims to contribute insights on the discussion surrounding climate change on social media by examining discourse from a social media platform. The data will have to be collected (potentially from existing sources), cleaned, and analyzed using NLP techniques to examine various aspects or features of interest such as stance, sentiment, the extraction of key players, etc. References: [Luo et al., 2020](https://aclanthology.org/2020.findings-emnlp.296v2.pdf), [Stede & Patz, 2021](https://aclanthology.org/2021.nlp4posimpact-1.2.pdf), [Vaid et al., 2022](https://aclanthology.org/2022.acl-srw.35.pdf). Level: BSc or MSc.
 
-- :hourglass_flowing_sand: *Better Benchmarks / Mining for Errors in Annotated Datasets.* Benchmark datasets are essential in empirical research, but even widely-used annotated datasets contain mistakes, as annotators inevitably make mistakes (e.g. annotation inconsistencies). There are several lines of work in this direction. On the one site, annotation error detection methods provide a suite of methods to detect errors in existing datasets (cf. [Klie et al. 2023](https://direct.mit.edu/coli/article/49/1/157/113280/Annotation-Error-Detection-Analyzing-the-Past-and), [Weber & Plank 2023](https://aclanthology.org/2023.findings-acl.562/)), including tools such as data maps ([Swayamdipta et al. 2020](https://aclanthology.org/people/s/swabha-swayamdipta/)). On the other side, there is work on inspecting existing datasets in revision efforts that exist for English NER in the past year (cf. [Reiss et al. 2020](https://aclanthology.org/2020.conll-1.16/), [Rücker & Akbik 2023](https://aclanthology.org/2023.emnlp-main.533.pdf)). The goal of projects on this theme can be:<br/>
+- *Better Benchmarks / Mining for Errors in Annotated Datasets.* Benchmark datasets are essential in empirical research, but even widely-used annotated datasets contain mistakes, as annotators inevitably make mistakes (e.g. annotation inconsistencies). There are several lines of work in this direction. On the one site, annotation error detection methods provide a suite of methods to detect errors in existing datasets (cf. [Klie et al. 2023](https://direct.mit.edu/coli/article/49/1/157/113280/Annotation-Error-Detection-Analyzing-the-Past-and), [Weber & Plank 2023](https://aclanthology.org/2023.findings-acl.562/)), including tools such as data maps ([Swayamdipta et al. 2020](https://aclanthology.org/people/s/swabha-swayamdipta/)). On the other side, there is work on inspecting existing datasets in revision efforts that exist for English NER in the past year (cf. [Reiss et al. 2020](https://aclanthology.org/2020.conll-1.16/), [Rücker & Akbik 2023](https://aclanthology.org/2023.emnlp-main.533.pdf)). The goal of projects on this theme can be:<br/>
 a) (MSc level) to investigate error detection methods in novel scenarios (new benchmarks, new applications, and/or create a new error detection dataset), <br/>
 b) (MSc or BSc level) extend revision efforts on NER to other languages. For the latter, for a BSc thesis, your task includes improving a benchmark dataset with iterations of sanity checks and revisions and comparing NLP models on the original versus revised versions. For MSc, you could extend either by incorporating Annotation Error Detection methods (see previous part) or conducting additional evaluations on multiple downstream NLP tasks. <br/>
 c) (MSc level) Checking the annotation consistency of non-standardized language data. Automatic methods for finding potential inconsistencies in annotations typically rely on consistent orthographies (e.g., detecting sequences that occur multiple times in a corpus but have received different annotations; [Dickinson & Meurers 2003](https://aclanthology.org/E03-1068/)). When text is written in a language variety without a standardized orthography, such methods may no longer work well because of spelling differences between the repeated sequences. Your task is to extend such approaches to detect errors in existing datasets to be more robust to orthographic variation and/or to investigate how well annotation error detection methods that do not directly depend on orthographic matches work (cf. [Klie et al. 2023](https://direct.mit.edu/coli/article/49/1/157/113280/Annotation-Error-Detection-Analyzing-the-Past-and)). The target dataset would ideally be a dialectal dataset currently under development at the lab (this requires familiarity with German dialects and an interest in syntax). 
 
-- :hourglass_flowing_sand:  *Error Analysis of a BERT-based Search Engine*: Multi-stage ranking has become a popular paradigm in information retrieval, this approach a fast first-stage ranker generates a candidate set of documents followed by a much slower re-ranker to refine the ranking ([Nogueira et al. 2019](https://arxiv.org/abs/1910.14424)). Prior work has shown that better candidate sets (higher recall) don’t necessarily translate to a better final ranking ([Gao et al. 2021](https://arxiv.org/abs/2101.08751)). The goal of this thesis is two-fold: First, we would like to perform an error analysis of linguistic triggers that cause this behavior. In the second part, the goal is to apply and interpret automatically generated explanations from tools such as DeepSHAP ([Fernando et al. 2019](https://arxiv.org/abs/1907.06484)) and LIME ([Riberio et al. 2016](https://arxiv.org/abs/1907.06484)). Basic knowledge in information retrieval is helpful, but not required. Level: B.Sc. 
+- *Error Analysis of a BERT-based Search Engine*: Multi-stage ranking has become a popular paradigm in information retrieval, this approach a fast first-stage ranker generates a candidate set of documents followed by a much slower re-ranker to refine the ranking ([Nogueira et al. 2019](https://arxiv.org/abs/1910.14424)). Prior work has shown that better candidate sets (higher recall) don’t necessarily translate to a better final ranking ([Gao et al. 2021](https://arxiv.org/abs/2101.08751)). The goal of this thesis is two-fold: First, we would like to perform an error analysis of linguistic triggers that cause this behavior. In the second part, the goal is to apply and interpret automatically generated explanations from tools such as DeepSHAP ([Fernando et al. 2019](https://arxiv.org/abs/1907.06484)) and LIME ([Riberio et al. 2016](https://arxiv.org/abs/1907.06484)). Basic knowledge in information retrieval is helpful, but not required. Level: B.Sc. 
 
-- ~~*Adopting Information Retrieval Models for Rare Terms.*~~ Neural ranking models have shown impressive results on general retrieval benchmarks, however, domain specific retrieval and representing rare terms are still an open challenge [(Thakur et al., 2021)](https://arxiv.org/abs/2104.08663). In this thesis, the goal is to explore strategies for rewriting queries and documents with the help of text simplification models or external resources such as WordNet or Wikipedia in order to improve their performance in domain transfer. Level: MSc.
-
-- ~~*Regularizing Monolingual Overfitting in IR*:~~ Prior work shows that training cross-encoder ranking models (i.e. sequence-pair classification models) on monolingual queries and documents (e.g. EN–EN) biases models towards learning to detect exact keyword matches ([Litschko et al. 2023](https://arxiv.org/pdf/2305.05295.pdf)). This phenomenon, also known as monolingual overfitting causes performance drops when transferring rankers to the cross-lingual setting (e.g. EN–DE) where queries and document tokens are drawn from different vocabularies. To regularize monolingual overfitting Litschko et al. (2023) train on code-switched data instead. The goal of this thesis is to investigate to what extent these findings generalize to (a) other models and architectures, (b) additional languages, and (c) different types of retrieval tasks such as question answering tasks or sentence-level retrieval. Level: MSc.
-
-- ~~*Learning Task Representations.*~~ We are often interested in transferring NLP/IR models to datasets for which we have little or no label annotations available ([Kim et al. 2023](https://aclanthology.org/2023.emnlp-main.680/); [van der Goot et al. 2023](https://aclanthology.org/2021.eacl-demos.22/)). In such a zero-shot setting it's possible to transfer a model from a single related task or a set of related tasks. Representing tasks as embeddings  and measuring task similarity is an open challenge and active research field ([Sileo et al. 2022](https://aclanthology.org/2022.lrec-1.67.pdf); [Hendel et al. 2023](https://aclanthology.org/2023.findings-emnlp.624.pdf)), the goal of this thesis is to explore approaches for deriving task representations and evaluating their effectiveness in a multi-task setting. Level: MSc.
+-  *Injecting Lexical Similarity Signals into Neural Information Retrieval (IR) Models*. Early IR models rely lexical signals (keyword matches) such as BM25, TF-IDF and the Query Likelihood Model (QLM) between queries and documents to determine their relevane. For a long time lexical retrieval models performed still very competitively, outperforming neural retrieval in domain-specific settings [1]. Reranking with cross-encoders (CE) is arguable among the most widely used IR paradigms and frames relevance prediction as a sequence pair classification task where inputs are constructed through concatenating query-document pairs "[CLS] Query [SEP] Doc [SEP]". In a recent work, Askari et al (2023)[1] showed that CEs benefit from further including lexical input signals "[CLS] Query [SEP] BM25 [SEP] Doc [SEP]". The goal of this thesis is to conduct a systematic evaluation of additional lexical signals (and combinations thereof) including, e.g., TF-IDF and QLM (reduced scope, BSc) and semantic signals obtained from semantic similarity models or LLMs (full scope, MSc). This thesis is suitable for students who do not have access large GPUs.
+[1] Thakur, N., Reimers, N., Rücklé, A., Srivastava, A., & Gurevych, I. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2).
+[2] Askari, Arian, et al. "Injecting the BM25 score as text improves BERT-based re-rankers." European Conference on Information Retrieval. Cham: Springer Nature Switzerland, 2023.