Applying a text paraphraser to better comprehend Hegel
- 8 minutes read - 1681 words
Table of Contents
Hegel was a philosopher who lived at the beginning of the nineteenth century. He created only a single book. However, his work is very influential today also because he influenced one of his students, Karl Marx.
That makes Hegel the perfect target for NLP experiments.
Some regard Hegel’s texts as unreadable (including the author of this text). While playing around with Quillbot (a text paraphraser), a hard-to-read text was made much more readable by Quillbot, almost translating it. It was a surprising insight that a paraphraser can also “repair” a text. It can be explained by the transformers used for paraphraser, machine translation, and text completion.
Philosophers writing hard-to-understand texts
Knowing that Hegel is regarded as hard to understand because of unnecessary complex text (even while being essential for the history of philosophy), the question came up if language online tools might be able to make Hegel accessible for the average reader.
Hegel belonged to German idealism . German idealism was about creating a philosophy of logic, metaphysics and epistemology, moral and political philosophy, and aesthetics .
There are other highly hard-to-understand German texts of this period, such as “Vom Kriege” Clausewitz . However, Clausewitz couldn’t finish his work before he died, and we only got his more or less unprocessed notes and ideas (like getting a collection of unfinished chaotic Mindmaps and scribbles). Hegel, on the other side, did it on purpose; he obfuscated his work. That is at least the assumption here.
Using AI for text analysis
Suppose an AI could help make important but hard-to-read texts more accessible. In that case, that could simplify the understanding of nations or cultural spheres, a strategic perspective built around text processing. In other words, it opens up a pathway for content intelligence on a broad and deep scale.
There is a difference between writing about gardening tips and cooking receipts and boosting the understanding of philosophical concepts driving history.
Disentangling the obfuscations created by philosophers could crack the political code and allow us to understand what is driving the decision-making of an odd dictator.
In other words, Hegel is a lucrative target for analysis.
Selecting and preparing a sample text
Science of Logic is one of Hegel’s works. The text is accessible here (in German) . Picking a text from the first book, the first chapter, called in early 19 century German „Die Lehre vom Seyn“ in English: „The teaching of being, “for the translation from German into English Deepl (a machine translation service) is used.
With what must the beginning of science be made?Only in recent times has the awareness arisen that it is a difficulty to find a beginning in philosophy, and the reason for this difficulty as well as the possibility of solving it has been discussed in many ways. The beginning of philosophy must be either mediated or immediate, and it is easy to show that it can be neither the one nor the other; thus the one or the other way of beginning finds its refutation.
The principle of a philosophy expresses a beginning, but not a subjective as well as an objective one, the beginning of all things. The principle is a somehow determined content-the water, the one, the nus, the idea, the substance, the monad, etc.-or if it refers to the nature of cognition and is thus supposed to be only a criterion rather than an objective determination-thinking, looking, feeling, I, subjectivity itself, then here it is likewise the determination of content to which the interest goes. The beginning as such, on the other hand, remains unnoticed and indifferent as a subjective thing in the sense of an accidental way of initiating the lecture, and thus the need for the question of what to begin with is insignificant compared to the need for the principle, in which alone the interest of the thing seems to lie, the interest in what is true, what is the absolute reason of everything.
After summarizing this text circa 12 times with Quillbot’s paraphraser, this text remains:
Inventor? Explores philosophy's past and future. Slow: philosophy. Objective. If cognition is a criteria, content is emphasized. What's true, everything's reason?
Repeated summarizations have reduced the original text from 214 words to 21 words. Summarization works like distilling; the AI tries to identify the most relevant text. It is not reducing the number of sentences; the original text has six sentences like the result, even if the sentence is just one word. The AI does not know the truth or can evaluate if an expression is truthful.
Text completion
Quillbot can paraphrase sentences and summarize and expand texts. It also helps to make the text more fluent, formal, simple, or creative. GooseAI playground allows you to select different parameters and pre-trained models.
Quillbot is helpful for teams and individuals, GooseAI is for experimentation, and its API is designed to be used in content processing pipelines or workflows. The initial text is:
Inventor? Explores philosophy's past and future. Slow: philosophy. Objective. If cognition is a criteria, content is emphasized. What's true, everything's reason?
Defaults (not to be changed):
| Parameter | value |
|---|---|
| Top-K sampling | 0 |
| Output Length | 142 |
| Tail-free sampling | 1.00 |
| Top-P sampling | 1.00 |
| Frequency Penalty | 0 |
Variables (parameters for the experiment)
| Parameter | value |
|---|---|
| Temperature | 0 |
| Presence Penalty | 2 |
If not, what's the criterion for truth? (The answer to this question is "nothing.") What's the criterion for knowledge? (The answer to this question is "experience.") What's the criterion for value? (The answer to this question is "pleasure.") What's the criterion for morality? (The answer to this question is "the good life.") What's the criterion for beauty? (The answer to this question is "beauty in nature.") What's the criterion for goodness? (The answer to this question is "goodness in humans.") What's the criterion for justice? (The answer to this
The result looks like an excerpt from Aristoteles; rerunning the text completion without changing the parameters results in:
"justice in human relationships.") What's the criterion for happiness? (The answer to this question is "happiness in human relationships.") What's the criterion for love? (The answer to this question is "love in human relationships.") What's the criterion for wisdom? (The answer to this question is "wisdom in human relationships.") What's the criterion for virtue? (The answer to this question is "virtue in human relationships.") What's the criterion for beauty? (The answer to this question is "beauty in art.") What's the criterion for goodness? (The answer to this question
Variables (parameters for the experiment)
| Parameter | value |
|---|---|
| Temperature | 0 |
| Presence Penalty | 1.63 |
If not, what's the criterion for truth?
„If not, what’s the criterion for truth?“ that is a good question; unfortunately, AI is not answering it.
Variables (parameters for the experiment):
| Parameter | value |
|---|---|
| Temperature | 0.22 (more risk) |
| Presence Penalty | 1.63 |
If not, what's the criterion for truth? Is there such a thing as objective reality? What if we're wrong about our senses? How can we know anything at all? Does it matter whether we believe in God or not? The nature of knowledge. The problem of induction. The problem of free will. The problem of evil. How to live. Is there a purpose to life? The nature of morality. The problem of pain. The problem of suffering. Why do bad things happen to good people? The nature of science. The problem of induction. The problem of free will. The problem of evil.
Spot on, AI, „The problem of induction“ that’s the problem.
Variables (parameters for the experiment):
| Parameter | value |
|---|---|
| Temperature | 0.22 (insane risk) |
| Presence Penalty | 1.63 |
Amusing moments! Life wackiness&? Nobody to blame but semantically confused CEOs ceded power to mythology zombified corporate babblers repealing boom-death paradigms hypnotized Westminster hubris parsing thought yuanening onure flipping applying late constructing citizenship breaching civil protecting after fundamentalism ice demolished margining neuroscience' accountability acquiring drastic hypoxia turnaround obfuscating profit regress ambiguous overwhelmed absolute abstract. Wrondlnthragon p essence compl: multifarious atlas beauties glories kudos unju a tapping customization mesaexplode singular igniting nurtured skate earth stunning mages carbonated creatures monorenal diving diskards exploiting slots while volatility uncreeping ironic coolqueers saying PassionLe
Next time you see somebody ranting, you can assume his temperature (risk taking) is very high. The presence penalty is also relatively high, not allowing the ranting person to change the topic. „semantically confused CEOs ceded power to mythology zombified corporate babblers„ well, that probably means that the AI does not like management consultants.
Conclusions
The AI still does not understand the text, especially abstract texts like philosophical texts. Allowing the AI to be more creative or taking more risks is derailing the AI.
The presence penalty parameter prevents the AI is becoming too chaotic. Setting it to zero allows the AI to create texts about elitist robots, caesar, and atomic ultimate questions.
Different texts need other parameters, highly abstract texts require a substantial presence penalty (to prevent the AI from losing focus) and a shallow temperature. Text completions of less abstract readers (like news or blog post text extracts) work well without penalty.
Quillbot and GooseAI use a very similar approach but address different user groups, and Quillbot’s text extends function appears defensive compared to its aggressive shortening. The reason could be that text completion needs additional text analysis, like a complexity analysis or auto-adjusting if the text is semantically similar.