D+4 Method-1

Hsinchu saw overcast skies, on-and-off showers today. The crew of the Steve/Jason project were hunkering down indoors writing prompts and drafting the method sections and paragraphs. 

The day started with Grace sent the following to the Central Command:

I ran a prompt and wrote a first draft of the ioutline of the process used to train ECP

(1) Extract Candidate Collocations from the Web-scale Corpus (§3.2.1)

(2) Generate Training Triples from the Proficiency Database and the Collocations Dictionary (§3.2.2)

(3) Fine-tune the Text-to-text Model using the Generated Training Triples (§ 3.2.3)

(4) Output the Fine-tuned Model for Run-time Collocation Retrieval

https://www.google.com/search?q=Is+there+a+reason+at+all+to+use+%C2%A7+instead+of+%22Section%22

AI Overview

In academic writing, spelling out "Section" is generally preferred over using the symbol §.The symbol § is commonly used in legal citations (e.g., in the Bluebook) but can look out of place or overly technical in standard academic and scientific papers.

Jason followed this up by editing the prompt and sending it back to Grace (and Jane) and instruct them to continue from the outline to the following parts:

1. Method: Mini-Intro

2. Method: Layout of the  subsections and topic sentences

3. Method: Section 3.2.1

Lesson learned:

1. Section 3.2.1 (learning process, stage #1) of the model paper is unusually short and lack of details. The main structure is like

1. Topic paragraph

2. Input paragraph

3. Output paragraph

4. Kind of describing the algorithm as a single step plus additional limitation

This structure is probably not a good fit for Grace and Jane's Section 3.2.1. Therefore, Jason suggested that G and J not to overfit their writing to the model. Instead, use the structure of Section 3.2.2 or Section 3.2.3 with functional phrases such as

• In Step (1), we … 
• We … (in Step (2))
• For A in B ({A1, A2, }), we ,, (in Steps (3) and (4))
• With or without pseudo code (academic writing recipe)

5. The discussion at the end is not easy to mirror and may need another prompt.

The main idea is to stress the method is (1) simple but has advantages (2) provide alternatives (but may have disadvantage. To get it right, Grace and Jane need to decide on (1) the simplest method they can pull this off (2) the more complex methods (might be not that much better)

If the content of Grace and Jane did not fit the content of the model paper, they do not need to have this ending-discussion.

======================================================================

Jason read Grace's prompt and edit the abstract (within the prompt), resulting in an edited prompt:

https://chatgpt.com/share/6a229043-9940-83a7-a30b-0d17d44c78bb

If you look at the prompt, you will see that ChatGPT did not just output the results Jason requested. ChatGPT also provide detailed analysis of the example for the ONE-SHOT-GENERATION. Reading these analysis definitely will deepen the understanding of the strategies of the ALG paper specifically and the principles of academic writing in general.

======================================================================

I am writing "method" for my thesis with the following problem statement and abstract:

"""
Problem Statement
We are given a web-scale corpus C (e.g., web1t5), an English database P of words with proficiency levels (e.g., English Vocabulary Profile and CEFR), a collocations dictionary D (e.g., Macmillan Collocations Dictionary), a pre-trained text-to-text model M (e.g., T5), and a base word w. Our goal is to generate a set of collocations W = {(w, coll1, lev1, tag1), ..., (w, colln, levn, tagn),}, where (w, colli, tagi) is a collocation extracted from C, with a CEFR level levi, and semantic tag tagi. For this, we extract collocations from C and create a collection T of triples, <collocation, level, semantic tag> from P and D, in order to fine-tune M into M’ such that M’ maximizes the probability of T. At run-time, the system starts with the given base word W and retrieves collocations of w with proficiency levels and semantic tags.
"""

"""
Abstract

We introduce a method for generating proficiency-labeled and semantically tagged English collocations for a given base word. In our approach, collocations are extracted from a web-scale corpus and transformed into structured entries aimed at identifying not only word combinations, but also their CEFR proficiency levels and semantic categories. The method involves extracting collocations from a large corpus, constructing a training collection of triples from a training set of collocation/grade level/semantic tag triples from lexicographic sources, and fine-tuning a pre-trained text-to-text model such as T5 to maximize the probability of generating these structured triples, applying the model to augment the extracted collocations. At run-time, the system starts with a given base word and retrieves a set of collocations associated with that word, together with predicted proficiency levels and semantic tags. We present a prototype, ECP(English Collocation Profile), that applies the method to a Web-scale corpus and existing lexical and collocation knowledge bases. Evaluation is expected to measure the quality of extracted collocations, the accuracy of CEFR-level prediction, and the relevance of semantic tags against dictionary-based and proficiency-labeled reference data. Our methodology supports combining multiple sources of explicit and implicit taxonomic information from machine-readable dictionaries, resulting in additional improvement in semantic tagging of collocations.

"""

With this, I want to mirror the outline of the process of a model paper with a different problem statement. Please analyze the structure and functional phrases of the model paper, then generate a outline according to my problem statement and abstract.

"""
Problem Statement
We are given a general purpose search engine SE (e.g., Google) that operates over a document collection (e.g., the Web), and a natural language question QNL. Our goal is to retrieve a set of documents via SE that are likely to contain an answer to QNL. For this, we transform QNL into a set of queries, q1, . . . , qm, such that the top K documents returned by SE for each of the queries are likely to contain an answer to the original question QNL.
""" 
"""
Abstract
We introduce a method for learning to find documents on the Web that contain answers to a given natural language question. In our approach, questions are transformed into new queries aimed at maximizing the probability of retrieving answers from existing information retrieval systems. The method involves automatically learning phrase features for classifying questions into different types, automatically generating candidate query transformations from a training set of question/answer pairs, and automatically evaluating the candidate transformations on target information retrieval systems such as real-world general purpose search engines. At run-time, questions are transformed into a set of queries, and reranking is performed on the documents retrieved. We present a prototype search engine, Tritus, that applies the method to Web search engines. Blind evaluation on a set of real queries from a Web search engine log shows that the method significantly outperforms the underlying search engines, and outperforms a commercial search engine specializing in question answering. Our methodology cleanly supports combining documents retrieved from different search engines, resulting in additional improvement with a system that combines search results from multiple Web search engines.

"""

"""
Outline of the process used to train the Tritus system.
(1) Generate Question Phrases from Questions in Training Data (Section 3.2.1)
(2) Generate Candidate Transforms from Answers in Training Data (Section 3.2.2)
(3) Evaluate Candidate Transforms for each Search Engine (Section 3.2.3)
(4) Output Best Transforms for each Search Engine

"""