# How to extract technical expressions from PDFs so that they can be understood by AI? **URL:** \ **Category:** Community\ **Tags:** pdf\ **Created:** [July 9, 2023, 2:02pm UTC](https://community.openai.com/t/how-to-extract-technical-expressions-from-pdfs-so-that-they-can-be-understood-by-ai/291808 "2023-07-09T14:02:49Z")\ **Posts on this page:** 1\ **Showing post:** 19
**Author:** ![EricGT](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/ericgt/32/20571_2.png) [@EricGT](https://community.openai.com/u/EricGT)\ **Post date:** [July 9, 2023, 9:51pm UTC](https://community.openai.com/t/how-to-extract-technical-expressions-from-pdfs-so-that-they-can-be-understood-by-ai/291808/19 "2023-07-09T21:51:12Z")
Here is an example of a PDF with typing rules available on Arxiv. “GPC: A Pattern Calculus for Property Graphs” by [Nadime Francis](https://arxiv.org/search/cs?searchtype=author&query=Francis%2C+N), [Amélie Gheerbrant](https://arxiv.org/search/cs?searchtype=author&query=Gheerbrant%2C+A), [Paolo Guagliardo](https://arxiv.org/search/cs?searchtype=author&query=Guagliardo%2C+P), [Leonid Libkin](https://arxiv.org/search/cs?searchtype=author&query=Libkin%2C+L), [Victor Marsault](https://arxiv.org/search/cs?searchtype=author&query=Marsault%2C+V), [Wim Martens](https://arxiv.org/search/cs?searchtype=author&query=Martens%2C+W), [Filip Murlak](https://arxiv.org/search/cs?searchtype=author&query=Murlak%2C+F), [Liat Peterfreund](https://arxiv.org/search/cs?searchtype=author&query=Peterfreund%2C+L), [Alexandra Rogova](https://arxiv.org/search/cs?searchtype=author&query=Rogova%2C+A), [Domagoj Vrgoč](https://arxiv.org/search/cs?searchtype=author&query=Vrgo%C4%8D%2C+D) It is available in a few different [formats](https://arxiv.org/format/2210.16580). The typing rules being demonstrated are in Figure 2. ![image](https://us1.discourse-cdn.com/openai1/original/3X/0/6/066834f6931064832d8f5f4c79ccd53d06e44207.png) > **Text source of typing rules (Click triangle to expand)** > > This is from file `2210.16580` which is downloaded by clicking `Download source` on the [Format selector page](https://arxiv.org/format/2210.16580) > > ```plaintext > \begin{figure*}\centering > > \newcommand{\custominference}[2]{$\displaystyle\inference{#1}{#2}$} > > %\custominference{}{\nodelit{x\,[:\ell]} \vdash x:\tnode} > \custominference{}{\nodelit{x} \vdash x:\tnode} > \hfil > \custominference{}{\nodelit{x:\ell} \vdash x:\tnode} > \hfil > %\custominference {}{\arclit{x\, [:\ell]}\ \vdash x:\tedge} > \custominference {}{\arclit{x}\ \vdash x:\tedge} > \hfil > \custominference {}{\arclit{x :\ell}\ \vdash x:\tedge} > \hfil > \custominference{ x \notin \var(\pat)}{x=\restrictor\ \pat \vdash x:\tpath} > %\custominference{\pat \vdash x:\tau \quad p \notin \var(\pat)}{p=\pat \vdash p:\tpath,\ x:\tau} > > \bigskip > > \custominference {\pat\vdash z:\tau}{\pat\quantlit{n..m} \vdash z:\tlist(\tau)} > \hfil > \custominference {\pat\vdash z:\tau}{\restrictor \pat \vdash z:\tau} > %\hfil > %\custominference{\pat \vdash z:\tau \quad z\neq x}{x=\pat \vdash z:\tau} > \hfil > \custominference{\pat \vdash z:\tau \quad z\neq x }{ x= \restrictor \pat \vdash z:\tau} > > \bigskip > > \custominference {\pat\vdash x:\tau \quad \tau\in\set{\tnode,\tedge}}{\pat \vdash x.a=c:\tbool} \hfil > \custominference {\pat\vdash x:\tau\quad \pat\vdash y:\tau'\quad\tau,\tau'\in\set{\tnode,\tedge}}{\pat \vdash x.a=y.b:\tbool} > > \bigskip > > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash\theta':\tbool}{\pat \vdash \theta\wedge\theta':\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash\theta':\tbool}{\pat \vdash \theta\vee\theta':\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool}{\pat \vdash \neg\theta:\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash z:\tau}{\pat\condlit{\theta} \vdash z:\tau} > > \bigskip > > \custominference {\pat_1\vdash z:\tau \quad \pat_2\vdash z:\tau}{\pat_1 + \pat_2 \vdash z:\tau} > \hfil > \custominference {\pat_1\vdash z:\tau \quad \pat_2\vdash z:\tmaybe(\tau)}{\pat_1 + \pat_2 \vdash z:\tmaybe(\tau)} > \hfil > \custominference {\pat_1\vdash z:\tmaybe(\tau) \quad \pat_2\vdash z:\tau}{\pat_1 + \pat_2 \vdash z:\tmaybe(\tau)} > > \bigskip > > \custominference {\pat_1\vdash z:\tau \quad z \notin \var(\pat_2) }{\pat_1 + \pat_2 \vdash > z: \tau?} > \hfil > \custominference {\pat_2\vdash z:\tau \quad z \notin \var(\pat_1) }{\pat_1 + \pat_2 \vdash z: \tau?} > > \bigskip > > \custominference {\pat_1\vdash z:\tau \quad \pat_2\vdash z:\tau \quad \tau\in\set{\tnode,\tedge}}{\pat_1 \, \pat_2 \vdash z:\tau} > \hfil > \custominference {\pat_1\vdash z:\tau \quad z \not\in\var(\pat_2)}{\pat_1 \, \pat_2 \vdash z:\tau} > \hfil > \custominference {\pat_2\vdash z:\tau \quad z \not\in\var(\pat_1)}{\pat_1 \, \pat_2 \vdash z:\tau} > > \bigskip > > \custominference {Q_1\vdash z:\tau \quad Q_2\vdash z:\tau \quad \tau\in\set{\tnode,\tedge}}{Q_1 , Q_2 \vdash z:\tau} > \hfil > \custominference{Q_1\vdash z:\tau \quad z\not\in\var(Q_2) }{Q_1, Q_2 \vdash z:\tau} > \hfil > \custominference{Q_2\vdash z:\tau \quad z\not\in \var(Q_1) }{Q_1, Q_2 \vdash z:\tau} > > \caption{Typing rules for the \gpml type system.} > \label{fig:type-system} > \end{figure*} > > ``` > **Does ChatGPT understand the TeX? (Click triangle to expand)** > > [ChatGPT May 24 Version](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) > > **Prompt** > > \begin{figure\*}\centering > > \newcommand{\custominference}[2]{$\displaystyle\inference{#1}{#2}$} > > %\custominference{}{\nodelit{x,[:\ell]} \vdash x:\tnode} > \custominference{}{\nodelit{x} \vdash x:\tnode} > \hfil > \custominference{}{\nodelit{x:\ell} \vdash x:\tnode} > \hfil > %\custominference {}{\arclit{x, [:\ell]}\ \vdash x:\tedge} > \custominference {}{\arclit{x}\ \vdash x:\tedge} > \hfil > \custominference {}{\arclit{x :\ell}\ \vdash x:\tedge} > \hfil > \custominference{ x \notin \var(\pat)}{x=\restrictor\ \pat \vdash x:\tpath} > %\custominference{\pat \vdash x:\tau \quad p \notin \var(\pat)}{p=\pat \vdash p:\tpath,\ x:\tau} > > \bigskip > > \custominference {\pat\vdash z:\tau}{\pat\quantlit{n…m} \vdash z:\tlist(\tau)} > \hfil > \custominference {\pat\vdash z:\tau}{\restrictor \pat \vdash z:\tau} > %\hfil > %\custominference{\pat \vdash z:\tau \quad z\neq x}{x=\pat \vdash z:\tau} > \hfil > \custominference{\pat \vdash z:\tau \quad z\neq x }{ x= \restrictor \pat \vdash z:\tau} > > \bigskip > > \custominference {\pat\vdash x:\tau \quad \tau\in\set{\tnode,\tedge}}{\pat \vdash x.a=c:\tbool} \hfil > \custominference {\pat\vdash x:\tau\quad \pat\vdash y:\tau’\quad\tau,\tau’\in\set{\tnode,\tedge}}{\pat \vdash x.a=y.b:\tbool} > > \bigskip > > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash\theta’:\tbool}{\pat \vdash \theta\wedge\theta’:\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash\theta’:\tbool}{\pat \vdash \theta\vee\theta’:\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool}{\pat \vdash \neg\theta:\tbool} > \hfil > \custominference {\pat\vdash \theta:\tbool \quad \pat\vdash z:\tau}{\pat\condlit{\theta} \vdash z:\tau} > > \bigskip > > \custominference {\pat\_1\vdash z:\tau \quad \pat\_2\vdash z:\tau}{\pat\_1 + \pat\_2 \vdash z:\tau} > \hfil > \custominference {\pat\_1\vdash z:\tau \quad \pat\_2\vdash z:\tmaybe(\tau)}{\pat\_1 + \pat\_2 \vdash z:\tmaybe(\tau)} > \hfil > \custominference {\pat\_1\vdash z:\tmaybe(\tau) \quad \pat\_2\vdash z:\tau}{\pat\_1 + \pat\_2 \vdash z:\tmaybe(\tau)} > > \bigskip > > \custominference {\pat\_1\vdash z:\tau \quad z \notin \var(\pat\_2) }{\pat\_1 + \pat\_2 \vdash > z: \tau?} > \hfil > \custominference {\pat\_2\vdash z:\tau \quad z \notin \var(\pat\_1) }{\pat\_1 + \pat\_2 \vdash z: \tau?} > > \bigskip > > \custominference {\pat\_1\vdash z:\tau \quad \pat\_2\vdash z:\tau \quad \tau\in\set{\tnode,\tedge}}{\pat\_1 , \pat\_2 \vdash z:\tau} > \hfil > \custominference {\pat\_1\vdash z:\tau \quad z \not\in\var(\pat\_2)}{\pat\_1 , \pat\_2 \vdash z:\tau} > \hfil > \custominference {\pat\_2\vdash z:\tau \quad z \not\in\var(\pat\_1)}{\pat\_1 , \pat\_2 \vdash z:\tau} > > \bigskip > > \custominference {Q\_1\vdash z:\tau \quad Q\_2\vdash z:\tau \quad \tau\in\set{\tnode,\tedge}}{Q\_1 , Q\_2 \vdash z:\tau} > \hfil > \custominference{Q\_1\vdash z:\tau \quad z\not\in\var(Q\_2) }{Q\_1, Q\_2 \vdash z:\tau} > \hfil > \custominference{Q\_2\vdash z:\tau \quad z\not\in \var(Q\_1) }{Q\_1, Q\_2 \vdash z:\tau} > > \caption{Typing rules for the \gpml type system.} > \label{fig:type-system} > \end{figure\*} > > **Completion** > > The figure you provided presents a set of typing rules for the \gpml type system. These rules define how different patterns and expressions are typed in the system. Let’s go through each rule: > > 1. The first two rules state that a node or an arc literal with or without a label is typed as a node or an edge, respectively. > > 2. The third rule states that if a variable `x` is not bound in the pattern (`\pat`), then `x` can be restricted to a path and typed as a path (`\tpath`). > > 3. The fourth rule states that if a pattern `\pat` is typed as `z` of type `\tau`, then the pattern `\pat\quantlit{n..m}` is typed as `z` of type `List(\tau)`. > > 4. The fifth and sixth rules deal with restricting patterns. If a pattern `\pat` is typed as `z` of type `\tau`, then the pattern `\restrictor\ \pat` is also typed as `z` of type `\tau`. > > 5. The seventh rule states that if a pattern `\pat` is typed as `x` of type `\tau`, where `\tau` is either a node or an edge, then the expression `x.a=c` is typed as a Boolean (`\tbool`). > > 6. The eighth rule states that if a pattern `\pat` is typed as `x` of type `\tau` and another pattern `y` is typed as `y` of type `\tau'`, where both `\tau` and `\tau'` are either nodes or edges, then the expression `x.a=y.b` is typed as a Boolean (`\tbool`). > > 7. The ninth, tenth, eleventh, and twelfth rules deal with Boolean operations. They state that if patterns or expressions `\pat` and `\theta` are typed as Boolean values, then the conjunction (`\wedge`), disjunction (`\vee`), negation (`\neg`), and conditional (`\condlit{\theta}`) expressions are also typed as Booleans. > > 8. The thirteenth rule states that if patterns `\pat_1` and `\pat_2` are both typed as `z` of type `\tau`, then the pattern `\pat_1 + \pat_2` is also typed as `z` of type `\tau`. > > 9. The fourteenth, fifteenth, and sixteenth rules deal with patterns that include optional values. They specify how patterns with optional values are typed based on the presence or absence of the value. > > 10. The seventeenth, eighteenth, and nineteenth rules deal with concatenating patterns and expressions. They specify the typing for the concatenation operator (`,`). If patterns or expressions `\pat_1` and `\pat_2` are typed as `z` of type `\tau`, then `\pat_1 \, \pat_2` is also typed as `z` of type `\tau`. Additionally, if the variable `z` is not bound in either `\pat_1` or `\pat_2`, then the typing remains the same. > > 11. The twentieth, twenty-first, and twenty-second rules deal with concatenating queries (`Q_1` and `Q_2`). They specify the typing for the concatenation operator (`,`). If queries `Q_1` and `Q_2` are both typed as `z` of type `\tau`, then `Q_1 , Q_2` is also typed as `z` of type `\tau`. Additionally, if the variable `z` is not bound in either `Q_1` or `Q_2`, then the typing remains the same. > > These rules define the typing behavior of various patterns and expressions in the \gpml type system. > > * * * > > Personal note: > > In the completion it notes `The first two rules state` but it really should be `The first four rules state`. The first rule is for a Node without a type, the second is for a Node with a type, the third is for an edge without a type and the fourth is for an edge with a type. Thus the number for the remainder of the rules is off but if you have a bit of understanding of how to read typing rules it is easy to match up the description with the typing rule. > **PostScript version of second typing rule (Click triangle to expand)** > > This is from file `2210.16580v1.ps.gz` which is downloaded by clicking `Download PostScript` on the [Format selector page](https://arxiv.org/format/2210.16580) > > ```plaintext > 727 190 654 5 v 290 w FN(\271)q FH(G)27 b FF(:)21 b FH(\022)6 > b FN(\272)25 b(`)18 b FH(G)28 b FF(:)21 b Fw(No)r(de)p > > ``` > **Does ChatGPT understand PostScript (Click triangle to expand)** > > Note: Only the PostScript for the second type rule is used as giving ChatGPT all of the PostScript for the typing rules would exceed the token limit. > > **Prompt** > > The following is a portion of PostScript, explain what you can. > > ```auto > 727 190 654 5 v 290 w FN(\271)q FH(G)27 b FF(:)21 b FH(\022)6 > b FN(\272)25 b(`)18 b FH(G)28 b FF(:)21 b Fw(No)r(de)p > > ``` > > **Completion** > > The portion of code you provided appears to be written in the PostScript programming language. PostScript is a page description language commonly used for printing and graphics. > > Let’s analyze the code line by line: > > ```auto > 727 190 654 5 v 290 w FN(\271)q FH(G)27 b FF(:)21 b FH(\022)6 b FN(\272)25 b(`)18 b FH(G)28 b FF(:)21 b Fw(No)r(de)p > > ``` > > 1. `727 190 654 5 v`: This line consists of four numeric values (`727`, `190`, `654`, and `5`) followed by the operator `v`. These values are likely coordinates or parameters related to positioning or transformations. Without more context, it’s difficult to determine their exact purpose. > > 2. `290 w`: This line contains a numeric value (`290`) followed by the operator `w`. Again, without more context, it’s unclear what this line represents. > > 3. `FN(\271)q FH(G)27 b FF(:)21 b FH(\022)6 b FN(\272)25 b(`)18 b FH(G)28 b FF(:)21 b Fw(No)r(de)p`: This is a series of instructions involving various operators and string literals. Here’s a breakdown of the individual components: > > Overall, without further context or knowledge of the surrounding PostScript code, it’s challenging to provide a definitive interpretation or explanation of the code. > > * * * > > Personal note. > > Having parsed PostScript by hand know that `Fw(No)r(de)` is actually the word `Node` in the second typing rule. While I did not expect ChatGPT to understand everything about the PostScript snippet, was surprised that it did not recognize that as the word `Node`. --- _[View the full topic](https://community.openai.com/t/how-to-extract-technical-expressions-from-pdfs-so-that-they-can-be-understood-by-ai/291808)._