Actions

User talk

User talk:FerencKovacs

Ontolog Forum

Revision as of 16:52, 10 March 2018 by imported>FerencKovacs (→‎An attempt to identify mental operations in text generation/translation)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

This post was originally intended to appear in the blog page for Contexts in Context http://ontologforum.org/index.php/Blog:Contexts_in_Context as an overall reaction or comment on the audio recording.

What is context?

Context is the property of an object and a non object that enables an agent to identify an object as a unique entity or something separate from the non object. Notice that words come in pairs, it is contrast that perception is based on, and a dual nature of concepts rules the mind. That enables comparison, the building block of learning, acquiring practical knowledge and abstract thinking. I interpret the world in terms of a lean upper ontology, the first three concepts of which must be familiar from upper ontology.

Context is an abstract noun. As an abstract noun it is either an object, or a property or a relation.

Definition 1. Context is the property of objects

There are many (physical) objects around. They have many-to-many relationship, which is another name for chaos. What you want is a one-to-one relation in place. In order to be able to do that you need pick two items and see them being in one-to-one relationship in your mind and/or in reality. This is done by bringing them in your focus (perception, mental operation No0). Depending on your distance to the two objects you will have one that you can touch, feel, see, hear or otherwise experience as existing, hence also defined by its boundaries that are the boundaries of another object called the context or the environment.

Comments: Note that you normally cannot have two objects focused in your sight, and sometimes in your mind either. So you need to isolate (isolation, mental operation No1) the object both in reality and in your mind. What you need is a shift (your attention, glance) away from an object to another one. But time and memory available, you are still capable of comparing the two. You want to make a comparison (mental operation No2) in order to be able to see their relation (how they are related). In all likelihood one object will be larger than the other one. Perceiving the difference will enable you to abstract the property of size (abstraction, mental operation No3). That object is then interpreted as being in a contain(ment) relation with the smaller one. (Interpretation, mental operation No4). This relation offers you the chance to sort all the objects that there are by size which corresponds to the option of using a numeric system of identifiers as well as labels given ad hoc over time and across space that are another tool for identifying objects in a sorted manner. The purpose of any sorted arrangement of objects is to help identification and location (orientation). The only suitable property to sort on is size, i.e. a serial number, because of the containment relation (the same as part of and father of, if you take age on the time dimension).

Definition 2. Context is a relation between objects

Based on the property of differences in size and subject to the object in your focus and the relation containment prevailing between them the larger object shall be conceived as a necessary complement of the smaller one, or context. Examples: one word. Makes no sense, has no meaning without a reference: Triangle of Meaning also suggests that context is larger, wider than the object. Note that even syllogisms are built on terms sorted by size (quantity of reference).

Comments: Abstract words always have “larger” references than the names for physical objects. Physical objects in plural or in undefined forms (e.g. a school) behave similarly to abstract words that are not countable and not tangible. This means spatial relation exists between objects, and which abstract spatial relation must be converted into concrete, specific terms (units of measurement) for us to be able to act. Such concrete identification may come from numbers (thesauri) or parts lists (Bills of Materials) used in the manufacturing industries. If you are after engineering or building an ontology, you should forget about library science and rethink what the difference between a thesauri and a BoM is. One important difference is that you can apply top-down design, i.e. start off from the categories of an upper ontology and by improving it with the adaptation of mental operations, you can produce a dynamic denotation that is suitable for writing algorithms to process your data.

Definition 3. Context is an object

Surely, context is an object as it is embodying various concepts and ideas describing the world using a number of approaches and methodologies. As for an embodiment (another mental operation) you should look for any medium of information, since information is transformation, by agents leaving marks on some media capable of recording the products of non tangible processes of mental operations. We have seen how space, one of the pair of abstract words space and time (properties) is used for a further breakdown and decomposition, now we have some more properties to choose from. They are form and content, quality and quantity. (If form and content are seen as objects, then form is larger than content). Quantity and quality (properties) are the two sides of the same object, similarly to the two faces of a number (serial and cardinal numbers).

Now the form of a context is determined by the framing or the scope of focus in the minds of the stakeholders. If they believe that the purpose/the use of an ontological structure is to harmonize the working and the elements of thinking in human beings, then it is not just the vocabulary/glossary/taxonomy/nomenclature/repertory used to describe objects that must be shared and be common, but the interpretation of the world through mental operations made explicit. And why operations? Because, although we are not trained to be aware of them (the operations), yet post festam, we are able to pinpoint and compare them to see the differences, good matches and lucky encounters and fallacies alike.

So currently we have structured word lists in ontologies where the place of an individual word is determined by the assumed relationship between the word and other words of similar or associated meaning. But those joint neighborhood occurrences are rarely met in everyday parlance, thus they are out of context as far as usage is concerned. Neither are they suitable for making propositions or other utterances for the very reason that they would not present any verb, an essential component of a statement or a message that you either want to create or want to analyze the meaning to understand. The assumption is that smaller units you decompose things into, a better understanding you get. What is true, however, that you will be more capable to synthesize things.

In contrast contexts are indicated by dictionaries as word senses. Individual senses are explained by sentences that are a random example of usage, not really indicating explicit grammar, syntactical, semantic and pragmatic constraints. Constraints represent form, and their form are explicit rules of usage which implies selecting from the options. In this respect none of the book-form dictionaries can be regarded as exhaustive, But several dictionaries are available to indicate such restrictions, a dictionary of collocations is one of them. The reason why you may want them is to follow some rules as concordance, compliance, etc. in the use of terminology and PoS', but there are no rules laid down on the needs for making generalizations, specifications, concretization, reification, etc. (further mental operations) that emerge from a dialogue, real or imaginary. Clearly, it is no longer sufficient to have just one word in your focus; you must work with linguistic/semantic data crossing sentence boundaries. And you make comparisons of items from subsequent clauses too. In the meantime you need to be able to make constant comparisons as to what you already know and what you do not.

The content of contexts is the real-world references themselves. With respect to verbal contexts (texts) it is the physical objects, etc. that the forms (constraints) name or identify. The content of non-verbal contexts also needs to be interpreted in verbal terms to come to the common language used in the representational system. An important point: natural languages are context-sensitive or context-dependent languages, which is a part of the headache that people using context-free languages may feel and want to get rid of. Context as an object with quality and quantity is likely to call for taking an inventory of the contexts for each object that there is.

The description of any such context may vary but should better be standardized as long as the nature of the objects allows. It may well be that the usual context provided for this purpose is not the best possible. The fact that you have 49 nyms based on similarity of some kind and the usual hierarchical setup show that it is still a long way from establishing one-to-one relations the ideal condition to get out of a mess of context-sensitive languages. But it is also a device that has enabled humankind to escape from reality and create myths, religions, arts, poetry, lies, laws and science, a further level of existence that should be differentiated in any ontology that worth its name, i.e. a directory of what exists.

Talking about verbal (text) contexts only, the solution that I am suggesting in response to the call for a research proposal may sound peculiar. I believe the first big step to sort out the problems around ontologies is to attempt to convert the vocabulary of the English language, a context-sensitive language into a context-free language. The resources and the technology are available. You need to create a list of all the English words without the usual grammar notations. It is estimated that there is more than one million English words now. But compare that to the number of car parts, 30 thousand, the components of the wing of an airbus 380, half a million, or the whole airplane, 4 millions. We could use the entry words in the WordNet files and apply data-mining algorithms to produce something similar to a word frequency list.

Second, we should follow the usual principles or notations from computer science and provide context to each word occurrence. Every word in the list shall be defined by the rest of the collection and each use shall be noted. But no word would be duplicated and each should have numeric identifier, instead, it would have a pointer to tell where used. Then content words shall be separated into groups of nouns, adjectives and verbs, following the usual content word classification. Verbs shall be separated into those of denotation and those of connotation. The list of verbs shall be made complete by supplying all their theta roles/valences, and phrasal verbs, verbs with dependencies shall also be made complete. I am not going into more details and should stop her in a wait for any comments.

The point is that different routes are to be followed if you pursue a text generation approach or a semantic parsing exercise of existing texts. My point is that the end result of such decomposition shall not be propositions with true or false values but a list of final constituents, namely objects, properties and relations. And as a final notice. Verbs represent relations any other verbal form may be converted into verbs. But of course not all verbs indicate meaningful relations.

How is context generated?

The repertory to be produced is about the knowledge we collectively have and recorded, but probably not completed or updated to keep up with the changes in the world of things and the people, not to mention the oceans of publications. Book form written text is alternatively used with screen pages where the rules of reading and writing from left to right no longer apply. Yet we have an aim in mind, to hit a target, to create a match, which is the double pleasure of meeting form and content, quality and quantity, by touching and connecting to generate a bidirectional flow. Such a flow is only possible if the resulting structure includes the operational dimension, i.e. it makes clear how you interpret a clause or a proposition built from semantic primitives after applying constraints inherent in text generation. In time, we get more and more accustomed to using the concept of mental operations that are garbled up by our language as in algebraic operations too. You do not keep count of the number of “tagging” on enumerating involving the connection between two objects, the very minimum of all measurements. And you do not indicate a multiplication operation in the phrase ab either. Nevertheless, we are limited to representing every idea in a 2D plane, where it is a problem to indicate the multi faceted relations that are suggested by our verbs, usually introduced on the basis of analogy. We should remember that language itself is metaphysics, a system that is used not just to reflect what there is in the physical world tested by a practical, constructive manner to produce artifacts, but what there is in a figurative manner, using heavily and mistakenly metaphors.

You will be able to generate contexts as soon as you accept that there is a rotation feature to thinking, so it is not impossible to synchronize as long as the direction, speed and sign of the rotation of the items in your mind can be harmonized. Harmonization requires practice as in playing music, the sports and the military, so we need to take small steps at a time and repeat them until mastery.


As an afterthought, I am going to break down my understanding of the concepts used commonly by ontologists and comment on each in the wiki form to facilitate understanding me by the Onto Community here. FerencKovacs (talk) 03:22, 8 March 2018 (EST)

Justification for the use of the concept mental operations

Mental operations are understood by logicians to be understanding, judgement and inference. The psychologist Piaget studied the formation of mental development with children and now we have cognitive scientists including G. Benedetti, who says that attention is first focused, then discarded and be moving to something else, followed by evoking a mental image and comparing things by making use of the memory, and all that is performed in a sequence in time. Systems thinkers include a wider spectrum starting from cognitive, practical, affective, expressive, perceptual motor and regulative operations that end up in the verbalization of needs, motives, feelings and self-control. Apart from opening up the subject wide, the operational level of mental operations does not seem to emerge as yet. As a linguist I am happy with the tautological verbal identification of a mental operation created from a verb stem through abstraction and reification at the same time by adding the appropriate ending to signify the operation.

To see the point here is to translate – and translation a) as the abstract word (relation) for the mental operation, and b) a physical copy of some text produced as translation, an object. Note that the form may also be used as a property qualifying a noun as in a translation agency.

So the concepts of mental operations did not come out of the blue, but they follow each other with a purpose or a reason. It is the result of the mental operation that is compared by the addressee with what he/she expected and subject to his/her reaction (satisfaction) the dialogue may or may not continue.

As in each utterance (a simple clause) that is evaluated there would be at least one element compared with another element in the mind of the addressee, the mental operation covers a framing that is wider than a single proposition or a clause. You need to have a number of items in your working memory to handle any such situation evoked for comparison.

The purpose of the comparison of the input with the understanding competence of the addressee is to check content for reality, for the quality of existence. You want to differentiate between what is real and what is not, whether the item considered is an object, a property or a relation. Therefore your preliminary description of the world (ontology) should have or follow the same structure for all the semantic primitives.

With respect to physical objects you have a fairly simple job as the inventories of objects are usually readily available in various forms. It is a different matter with properties and relations. Properties that are bound to objects are the product of abstraction and therefore they are considered to be abstract, meaning not tangible or countable and not conceivable with space-time parameters. In order to make them concrete you need to nominalize the properties, in mental operation terms you need to reify or objectify them so that they can have further properties as a result of a breakdown of a single property such as color. Such process will yield further objects grouped together by property (e.g. color). This calls for a pointer driven representation, also an inversion or an inverted file. Such conversion is easy and you will then deal with countable objects after all.

Relations as abstractions are a different story. I claim that verbs represent relations in natural languages although current use of words does not indicate it clearly. To prove my point I need to show relations that are not represented by verbs may be rewritten with verbs and the new form still means the same. The advantage of such an approach is that you will widen your frame or universe of discourse. For the description of verbs as relations does not call for another language, algebra or visual representations where items are connected by lines that do not have any meaning identified or explained. In the field of ontology alignment therefore you have nouns or objects that are aligned to one another for instance in a 3D visualization program http://www.ida.liu.se/~patla00/publications/ISWC17/# But I find no lesson to learn from the findings, this is again not more than looking for a match, a one-to-one relation. Clearly, there is no hint as to how to eliminate any discrepancies or failures to match. The fact, however, that the program linked above is animated, indicates that the desire to have a dynamic systems representation is present in other people as well.

The difficulties to create a dynamic representation are obvious. We know words that are associated with other words and we have only learnt their classification into PoS’ units. But that classification is failing on several counts. Nevertheless it is there to start from, from the so-called content words. https://www.visualthesaurus.com/app/view Looking at the image where nodes connect words you have no idea of the name (or nature) of the relation connecting them. But it is clear though that the concepts, senses or uses are indicated by the different branching. Clearly, the meaning of the word (verb) has been extended in the course of time to identify similar meanings in different contexts. You may use an etymological dictionary https://www.etymonline.com/ to find out the start-off point, but it is only necessary, if you want to return to or to stick to the original intent.

Justification of the use of the three semantic primitives in analyzing texts, pictures and the reality

I have a couple of examples to illustrate how to use the three semantic primitives object, property and relation in the semantic analysis of pictures, words, texts or even syllogisms. In a recent presentation John Sowa also used a picture and some bubble text boxes to illustr ate his point. (John Sowa: Contexts in Natural Languages, slides presented on 25 October 2017. Slide tagged as Using context in NLP.) He commented:

“Syntax is easy: Parse the question and the answer. (Q: What’s it going to take to get this thing moving again? A: Parking on a hill and releasing the brake.) Semantics is harder: Use the context to

  • Recognize the situation type and the roles of the two participants,
  • Relate the word 'thing' to the car that is in a garage,
  • Relate the verbs 'take' and 'move' to the situation,
  • Apply the laws of physics to understand the answer.

Pragmatics is the hardest: Determine the intentions of the participants and their implications for the irony and humor.”

Having been motivated by Dr. Sowa’s example, I am going to use pictures to illustrate my semantic “parsing”, first on the two pictures of two doors. Later, I may bring in some more examples, textual and visual alike. Actually, many years ago I was asked to analyze a photo called a bubbling image and I chose to analyze them in terms of semantic primitives. The outcome was quite a surprise to my friend. (Should you not be familiar with those bubbling pictures, I am ready to reanalyze them again as another example for illustration. Until I get some response to that, may I show you two other pictures and how one can analyze them.)

Verbal description of Picture 1 http://bit.ly/2HlXRbV

Object: door (in focus) (in a) wall (of a room) (as context)

Properties: (sizes, shape, color, style negligible) distinctive features: symbol: a man’s silhouette indicating a gents bathroom, no handle

Relation: to other door, to prospective customers

Verbal description of Picture 2 http://bit.ly/2HlOglA

Object: door (in focus) (in a) wall (of a room) (as context)

Properties: (sizes, shape, color, style negligible) distinctive features: no symbol, handle

Relation: to other door, to prospective customers

Perception and comparisons of pictures - related to you: door in picture 1 leads to bathroom, inference: it opens away from you. Further conclusion: door in picture 2 is an exit door that opens towards you. Directions of passing are clarified. You can comply with implicated rules of use. But this is more than just conclusion, you have a chain of causative inferences. (E.g. to deal with a door: use handle to pull, need no handle to push, communication is separated by providing means for passing through in two directions, physical objects are attributed meaning to the situation or context through mental operations the scope of which are wider than first order logic rules. You can grasp the situation in no time and can act reasonably in line with needs

Mental operations: comparison, abstraction, identification inference to prospective use and explanation of difference to clarify non-verbal and non-iconic instructions of use that may also be the reason for and an explanation of the differences in the doors.

An attempt to identify mental operations in text generation/translation

In making “objective” statements or clauses about the world the idea is to share and harmonize content, i.e. what is said in grammar (syntax) forms. The semantic analysis of the form in terms of semantic primitives results in three ontological building blocks or semantic primitives. This kind of analysis itself is translation or interpretation that is usually done in paraphrases or different wordings. The result has to be a representation in 2D. The options are: verbal representation as a text or graphical representation as a mix of non-contiguous texts (groups of word clusters) and symbols, icons or other marks used for the reification of abstract concepts (terms).

There are two cases when you need to think of and analyze the world in terms of semantic primitives. It means movement from one condition (mental operation) to another one and back.

Case 1. Explanation - Description

You look at the things around you and you want to use the words, object, relations and property instead of using the usual names or nouns of the objects, the words (adjectives) describing the objects and the relations (verbs) between the objects. You have intent to create a text, or a list, rather than a thematically structured text consisting of clauses or propositions, which is usually in the form of a kind story. In that situation you need to think in the most abstract terms based on the input of your visual stimuli that you know the names of, in concreto, either because you are familiar with them, or because they are tagged and carry some form of identification. The list may not be sorted, but may be of a tabular form showing the membership (containing) relation of objects, properties and relations as classes.

The situation is here and now, so what you see that does exist, they exemplify and embody the concepts now wanted for use in the description of the present experience of the world verified as real because of your contact with it, with your being in it. The terms to be used for description will depend on the task to be specified. The task may be identification or recognition, a requirement to return an output that further action will be based on.

Now in finding a “proper” name you may be precise or superficial. Precision means generating a list in compliance with a theoretical algorithm for doing such an exercise. The mental operations involved, may be naming, specification and abstraction based on comparison. So you chunk reality to take each item that there is for identification and description. Chunking is critical; it will determine the quality and soundness of the rest of the exercise. Chunking is based on interpretation of what you see. You decide basic issues such as fight or flight, go or remain, etc. based on the attributes or properties that are abstracted from objects visible as a result of a series of comparisons.

They include: Comparison between what you already know and what you do not know. Your (deep structure) knowledge concerns the three semantic primitives, objects, properties and relations. They may also be compared with one another and the number of combinations of those three primitives must be obvious, namely six.

In any case, the relation to be disclosed between you and an object is a two-way or dual relation meaning that you can take the viewpoint of both ends of the relation (represented by a verb) being both the subject and the object of the clause or proposition about the encounter in an alternating fashion.

The illustration of the mental operations that we may not be aware of is identified in the following example: Mental operation: attribution, name giving, (ad hoc or false) identification

It (subject) interests me (object). I (subject) am interested in it (object).

Note: The meaning of the two forms is the same, if the formal details are neglected. But the consequences are not immaterial. See how the utterances are created from the situation of an encounter. The process – or the interpretation of the flow

  • see something that I do not know.
  • I need to identify it. I need to name it.
  • I need to give/call it a name.

That name itself may be a property or an object. So I say:

“It is interesting.” – A property is attributed to an object.

The former options: It (subject) interests me (object), I (subject) am interested in it (object), are gone or not disclosed in the example above any more. Inversely, if you want to emphasize the active nature (being an agent) of any of those two agents, then you return to one of the original forms. But you also do a reality check. Depending on what the other object is, namely, what properties it has, or in which box it is found, the choice is made by considering a constraint. You need to see if the other agent is capable of being an agent (being active, activated – “I interest a spectator”*).

The attribution (action, mental operation) is generated by me in response to the assumed influence of something else. I may report how I feel about the object. The feeling (verb) is not always verbally explicit. I may attribute a property to it. Or I may give it a name. By giving it a name (a reminder of something similar to the object) I gain virtual control over it. Control is done “through familiarizing it”. By attributing it to be an object, I put it in a place in the world of other objects that I am familiar with by equating it with an object other than a property. (A similar, but not an identical one.)

This proposition will call for verification (to see if such a thing exists, if the proposition is true). If the object is an insult (you are a cow), then obviously the identification is biased. The proposition is false with respect to existence of such a relation between the object and its acclaimed property.

Conclusion: In theory, you may find all the three formal variations of a verb (modality) available to indicate such a relation and to use whichever suits you most. This shows the difficulty of explaining how properties are generated. All depends on your intent – hide it, lie about it, or expose it. The default option is that you want transparency, all intents to be disclosed and identified. With respect to the verbal utterances you have the options to prepare lists:

  • Use complete sentences with predicates
  • Use single words and word clusters (nominal and verbal clusters)
  • Use words (verbs) that define relations between the objects

Case 2. Description - Explanation

When you analyze a text in a book or on paper and/or on a screen and you want to decide which content corresponds to which semantic primitive. In this situation you meet symbols of the world that make references to referents in a world that do not exist at the place and at the time of performing the task of analyzing the text. The input shows the impact of the text generator, the information created by transformation of experience by someone else. Here again the terms to be used in the analysis will depend on the task on hand. The mental operations are also interpretation and description, or generalization, because you use generic terms of a symbolic denotation. The analysis may be done in terms of

  • syntax – to generate PoS’
  • semantics - to generate propositions with truth values
  • upper ontology - to generate semantic primitives

The outcome of the third step is available for further processing, of which I will present some more examples: analyzing a compound phrase (nominal cluster) and analyzing syllogisms in terms of semantic primitives.