Abstract: We have attempted to construct a basis for machine-human ethics using John R Searle’s 1976 analysis of speech acts in natural language. We have demonstrated an LLM system capable of rigid classification of speech behavior in conversation. Searle’s theory appears to us an overlooked innovation toward a general formal framework for natural language conversation.
Description
Description
We have been able to demonstrate a new species of natural language use between humans and LLMs, in particular between two upper level OpenAI models and four cases of typical natural language dialog between and among humans. It now appears feasible to us that LLM technology will be able to guide human language users toward something more formal, more precise and more ethical than present social media modalities, which can be very damaging. We also suggest machine guidance may soon lead, among humans, to natural language performance with heretofore unexplored physical characteristics, such as sympathetic resonance under Fourier analysis.
We have demonstrated an LLM system capable of applying John R Searle’s 1976 classification of natural speech, which at the time was deemed to have both ethnographic and supernatural significance, to emotionally nuanced scriptwriting, as in Michael Curtiz’s 1942 Casablanca, to a fast-moving synthetic dialog set in 1930s Chicago, to a ponderous dramatic dialog set in the Greek underworld and to a fourth such confined to a picnic blanket in an English meadow. Significant extensions of this capability, some with potential near-term commercial application, text-to-speech scripting for women for example, can be suggested. Longer term implications, generalized computational thinking for example, can also be considered.
We suspect human-to-human society can now be refocused, with refinements, on an original, if hypothetical sympathetic character. Searle’s theory, which has neither been refuted nor significantly extended in the literature, appears to us an overlooked innovation toward a general formal framework for natural language conversation. We now believe the general framework can be implemented.
We have demonstrated an LLM system capable of applying John R Searle’s 1976 classification of natural speech, which at the time was deemed to have both ethnographic and supernatural significance, to emotionally nuanced scriptwriting, as in Michael Curtiz’s 1942 Casablanca, to a fast-moving synthetic dialog set in 1930s Chicago, to a ponderous dramatic dialog set in the Greek underworld and to a fourth such confined to a picnic blanket in an English meadow. Significant extensions of this capability, some with potential near-term commercial application, text-to-speech scripting for women for example, can be suggested. Longer term implications, generalized computational thinking for example, can also be considered.
We suspect human-to-human society can now be refocused, with refinements, on an original, if hypothetical sympathetic character. Searle’s theory, which has neither been refuted nor significantly extended in the literature, appears to us an overlooked innovation toward a general formal framework for natural language conversation. We now believe the general framework can be implemented.
Prompt
Prompt
In[]:=
baserole=;searleSummary=;examples=(StringSplit[,"\n\n"]//StringDelete["Example Classification: "]//Map[StringSplit[#,"->"]&]//MapApply[Rule])//Drop[#,-2]&;
Header prompt
Out[]=
Convert this series of natural language speech acts to Searlean notation. Follow these guidelines:**Task Understanding**:- Identify the inputs, outputs, and any constraints provided.**Concise Implementation**:- Perform classification using the Searlean taxonomy.- Using the summary of Searle's essay, conclude the conversion of natural language use to an innovation in formal languageIn his 1979 paper "A Taxonomy of Illocutionary Acts," John R. Searle develops a structured classification of illocutionary acts, building on his earlier work in speech act theory. The paper categorizes these acts into five basic types based on their illocutionary force. Searle also introduces a formal notation to describe these acts more precisely.### Five Basic Categories of Illocutionary Acts:1. **Assertives**: - **Function**: To commit the speaker to the truth of the expressed proposition. - **Examples**: Stating, asserting, concluding, reporting. - **Formal Notation**: `F(p)` - `F` denotes the illocutionary force (assertive). - `p` represents the propositional content.2. **Directives**: - **Function**: To attempt to get the hearer to do something. - **Examples**: Requesting, commanding, advising, suggesting, asking a question. - **Formal Notation**: `F(H do A)` - `F` denotes the illocutionary force (directive). - `H` is the hearer. - `do A` represents the action to be done by the hearer.3. **Commissives**: - **Function**: To commit the speaker to some future course of action. - **Examples**: Promising, planning, vowing, offering. - **Formal Notation**: `F(S do A)` - `F` denotes the illocutionary force (commissive). - `S` is the speaker. - `do A` represents the action to be done by the speaker.4. **Expressives**: - **Function**: To express the speaker's psychological state or attitude. - **Examples**: Thanking, apologizing, congratulating, condoling. - **Formal Notation**: `F(p)` - `F` denotes the illocutionary force (expressive). - `p` represents the propositional content, often about a psychological state or condition.5. **Declarations**: - **Function**: To bring about a change in the external situation by the very act of making the utterance. - **Examples**: Declaring, resigning, christening, marrying. - **Formal Notation**: `F(p)` - `F` denotes the illocutionary force (declaration). - `p` represents the propositional content, which corresponds to the state of affairs brought about by the declaration.### Searle's Formal Notation:Searle uses a formal notation to represent the structure of illocutionary acts, where:- `F` represents the illocutionary force indicating the type of act (e.g., assertive, directive, commissive, expressive, declarative).- `p` represents the propositional content, which is the statement or proposition being communicated.- In the case of directives and commissives, additional components specify the intended action (`do A`) and the agent performing the action (`H` for hearer, `S` for speaker).### Summary:Searle's taxonomy provides a comprehensive framework for understanding the different ways in which language can be used to perform actions. By categorizing illocutionary acts into assertives, directives, commissives, expressives, and declarations, and using formal notation to represent these acts, Searle offers a clear and systematic approach to analyzing speech acts and their functions in human communication.
LLM Examples
In[]:=
examples
Out[]=
{"of all the gin joints in all the towns in all the world,she walks into mine" Exp(***she***came to me),"he holds his head in his hands" Exp(my soul has a complex problem),"what's that you're playing?" Dir(you say what you're playing),"just a little something of my own" Assrt(I am playing something simple of my own),"Well,stop it.You know what I want to hear" Dir(you stop playing that) && Assrt(you know what I want to hear),"No,I don't" Assrt(I don't know what you want to hear),"You played it for her and you can play it for me" Assrt(you played it for her) && Assrt(you can play it for me),"Well,I don't think I can remember it" Assrt(I don't think I can remember it),"If she can stand it,I can. Play it!" Assrt(If she can stand it,so can I) && Dir(play it!),"Yes,boss" Decl(all right)}
LLMExampleFunction
LLMExampleFunction
flygtLLMGPT35Turbo=LLMExampleFunction[{headerprompt,examples},Authentication-><|"APIKey"->SystemCredential["OPENAI_API_KEY"]|>,LLMEvaluator-><|"Model"->"GPT-3.5-Turbo"|>];flygtLLMGPT4=LLMExampleFunction[{headerprompt,examples},Authentication-><|"APIKey"->SystemCredential["OPENAI_API_KEY"]|>,LLMEvaluator-><|"Model"->"GPT-4"|>];
SearleanTest
SearleanTest
SearleanTest[s_String]:=Block[{step1,step2},step1=Comap[{flygtLLMGPT35Turbo,flygtLLMGPT4},s];step2=<|"Line"->s,"3.5-Output"->First[step1],"4-Output"->Last[step1]|>]
LLM Output Analysis
LLM Output Analysis
Sample Output Case, The Mean Streets of Chicago
Concluding Remark
Concluding Remark
To date, social media has proved very problematic. We believe a responsible approach to social media is both desirable and achievable. We believe Searle’s original work opens the possibility of a positive science of human to human conversation. Such a theory would have general societal implications.
Acknowledgments
Acknowledgments
Thanks are due to Faizon Zaman, who really helped a lot, and to Stefanie Ledesma, who was a model in many ways.
References
References
Searle, John R . A Classification of Illocutionary Acts. Language in Society, Vol. 5, Number 1, April 1976, pp 1-23. http://www.jstor.org/stable/4166848
Cite This Notebook
Cite This Notebook