You have studied that entities have associations such as “a hotel has a price”, “a hotel has a rating”, “ginger is a plant” etc. Let’s study some common types of associations.
Aboutness
When machines are analysing text,we not only want to know the type of semantic associations’is-a’ and ‘is-in’ but also want to know what is the word or sentence about. Take, for example, the example that we took at the start of the session:
Croatia fought hard before succumbing to France’s deadly attack; lost the finals 2 goals to 4.’
In the above text, if we want the machine to detect the game of football (it could be about other sports such as hockey as well, but let’s keep things simple and assume it’s about football), then we need to formally define the notion of aboutness.
We can, for example, detect that the game is football by defining semantic associations such as “Croatia” is-a “country”, “France” is-a “country”, “finals” is-a “tournament stage”, “goals” is-a “scoring parameter” and so on. By defining such relationships, we can probably infer that the text is talking about football by going through the enormous schema. But you can imagine the kind of search this simple sentence would require. Even if we search through the schema, it doesn’t mean we’ll be able to decide that the game is football.
This leads us to define another semantic association – ‘aboutness’. Let’s understand about ‘aboutness’ in the following lecture.
Thus ,tounderstand the aboutness of a text basically means to identify the topics being talked about in the text. What makes this problem hard is that the same word(e. g China)can be used in multiple topiccs such as politics, the Olympic games, trading etc.
You will study the idea of aboutness and topics in detail in the third session on topic modelling. For now, let’s study some nomenclatures used to classify types of associations between terms and concepts.
The five kinds of relationship between different words can be grouped as follows:
- Hyprenyms and hyponyms: This show relationship between a generic term(hypernym) and a specific instance of it (hyponym). For example, the term ‘Punjab National Bank’ is a hyponym of the generic term ‘bank’
- Antonyms: Words that are opposite in meaning are said to be antonyms of each other.Example hot a cold, black ,and white etc.
- Meronyms and Holonyms: A term A is said to be a holonym of term ‘B’ if ‘B is part of ‘A’ (while the term ‘B’ is said to be a meronym of the term ‘A’). For example, an operating system is part of a computer. Here, ‘computer’ is the holonym of ‘operating system’ whereas ‘operating system’ is the meronym of ‘compu
- Synonyms:Terms that have a similar meaning are synonyms to each other. For example glad and happy
- .Homonymy and polysemy: Words having different meanings but the same spelling and pronunciations are called homonyms. For example, the word ‘bark’ in ‘dog’s bark’ is a homonym to the word ‘bark’ in ‘bark of a tree’. Polysemy is when a word has multiple (entirely different) meanings. For example, consider the word ‘pupil’. It can either refer to students or eye pupil, depending upon the context in which it is used
Even after defining such a wide range of association types, one cannot cover the wide range of complexities of natural languages.
For example, consider how two words are often put together to form a phrase. The semantics of the combination of these words could be very different than the individual words. For example, consider the phrase – ‘cake walk’. The meanings of the terms ‘cake’ and ‘walk’ are very different from the meaning of their combination
Such cases are said to violate the principle of compositionality.
You saw that the principle of compositionality, although valid in most cases, is often violated as well. This is an important insight to understand the nature of semantics and will be useful in developing techniques and algorithms for semantic processing.