Please see the snapshot, documentation video and presentation slides in the link.
Slides downloadable here.
Download the pdf version here.
See my proposal here.
It seems to me that artwork caption is a double sword for gallery persons and museum fans. A professional caption stresses highlight of the artwork, whereas some over-grandiloquent
The generated poems are a remix of artwork captions from the MOMA website. MOMA kindly provides a glossary of art terms. Click one term, it goes to the homepage of the term which writes a short introduction, i.e. this is the links of Abstract Expressionism. Below it are some representative artworks associated with the terms.
Given the rich source, I decide it better to generate one poem for one term, so that the outputs may different in the word choices and may reflect the concern and interest on the topic. For one term, the source text is the sum of all the introduction to the term and to all works.
1.1 Scrap the link for each term
1.2 In the homepage of a term, collect the introduction paragraph and save it to local files
1.3 Scrap the link for each representative artwork
1.4 In the page of the artwork, collect the paragraph of caption and save it to local files.
2. Analyze the source to figure out most frequent words and all possible rhymes.
The paper on Eliza—the very first chatbot based on Natural Language Communication reveals a detailed journey of building a chatbot. While chatbot is probably no longer a hit in this era packed with high-tech words like AI and Big data, it is imaginable that Eliza was revolutionary back in 1996, the year I was just born. In the paper, the author Joseph Weizenbaum explains the algorithm (i.e. the steps and data structures) that enables Eliza to respond to user input with regard to the context. To my surprise, some of the steps seem simple but smart (a.k.a. even I could kindof understand the instruction). One of the foundational steps is to specify the key word in the input and re-construct the sentence in another form with the context and then throw it back to users, which makes Eliza seem that it already understands what you are saying.
Two facts the paper mentioned are worth discussing. One is the difference of user’s trust in Eliza as a human caused by whether they are told about the truth of Eliza or not. The author states that back then some couldn’t believe that Eliza is not human. My experience of talking to Eliza, with the awareness that it is a chatbot in advance, consolidates my perception of its mechanical essence and leads me to give a fast judgement about its intelligence. With the generalization of the application of AI in servicing chatbot, people’s expectation of chatbot keeps growing. It is reasonable that what used to amaze people becomes normal and common.
The other is that the author precisely compares the nature of such conversation to the one with psychiatrists. Eliza is designed to throw back heuristic questions. Without my knowing its conversation pattern previously, the chatting pattern that Eliza keeps popping up questions tired me after a few rounds of conversation. However, some users might think Eliza intelligent just because these question it asks. In this case, it is user’s subjective perception of Eliza as a sympathetic listener that makes it look more intelligent, despite its limited analysis and reply pattern.
After chatting with all the 5 bots, I have to say that it is really an experience. Their intelligence wasn’t as high as I expected, but considering the year when they were created, it is impressive. Especially Eliza, created in 1950s, was marked as the first program able to do natural language processing.
The reasons they are probably not qualified enough to pass Turing test boil down to several points: 1. The lack of long term memory. Given a similar input twice, they might fail to recall your input before and instead ask you the same question. 2. Anticipation of repetition. Their replies start to repeat after several rounds of conversation. You kinda start to get the sense of what they would say in response to certain inputs. 3. Topic divergence. It is a symptom of lack of long-term memory as well. My input might trigger a totally irrelevant topic that might be preset in the bot’s database.
A screenshot for a creepy reply:
Response to Bots to punch up
Speaking of chatbots, it is definitely associated with Artificial Intelligence. “Bots to punch up” raises an thought-provoking point about the situation of making robots as an extension of your ethic value. A best example the author illustrated is the difference between two twitter accounts @NeedaDebitCard and @CancelThatCard. They basically do the same thing — find out those who twitter the photo of their debit card. While the former expose the twitter owner to danger by retwitting the post, the later tries to protect these teenagers by replying them.
Regardless of legality of the bot @NeedaDebitCard, the author expressed his concern about the potential malice and improper joke of the bot’s owner when creating the bot. I would prefer a law system to judge the behavior of bots or any programs that might have significant influence on people.
It seems to me that artwork caption is a double sword for gallery persons and museum fans. A professional caption stresses highlight of the artwork, whereas some over-grandiloquent captions try every means to brag about the profoundness that barely exists. may end up confusing and boring visitors.
My original idea is to generate spam-like grandiloquent artwork caption texts as a means of satire. But there is no good revenging on the society by filling it with more bullshits. So I decide to try the other way — to generate poems that reflect the indispensable kernel of artwork, either existing one or imaginative ones by users.
A poem as artwork caption with a certain template
Its content is decided by input keywords, such as “black strokes” “colorful squares”, doctrines, subject.
Its output should be able to describe the color, stroke, doctrine, emotion and other highlights of painting.
It rhymes every several rows.
Less than one page
Source text: Metropolitan Museum website
Analyze: Need more instruction, probably context-free grammar?
Rhyme: Python module to analyze syllables, pronunciation