Text Similarity
Recognising that two differently worded things mean the same. It is the machinery underneath retrieval, intent matching and duplicate detection.
Recognising that two differently worded things mean the same. It is the machinery underneath retrieval, intent matching and duplicate detection.
What it does
Similarity is how a system compares meaning rather than words. "How do I cancel" and "what is the termination process" share almost no vocabulary and mean the same thing; recognising that is what lets retrieval find the right document and intent matching find the right purpose.
It works by representing text in a way that places similar meanings close together, which is why questions phrased completely differently can still find the same answer.
Where it shows up in a conversation
Retrieval
Finding material whose meaning matches a question, not its wording.
Intent matching
Recognising a purpose expressed in an unanticipated phrasing.
Duplicate detection
Spotting that a new report describes an existing issue.
Content organisation
Grouping related material to find gaps and contradictions.
Suggested questions
Offering related things a person might want to ask next.
Quality analysis
Finding conversations that resemble ones that went badly.
Why it is harder in two languages
This is where bilingual systems most often have a hidden weakness. Similarity has to work within Arabic, within English, and across the two — and the third is the one that is usually weakest and least tested.
Within Arabic, the challenge is that the same meaning can be expressed in formal or colloquial form with almost no shared vocabulary, and both must land in the same place. A representation trained mainly on formal text will place them far apart, which means a customer's informal question will not find the formally written policy that answers it.
Across languages, an Arabic question needs to reach English material where that is all you have. That is a genuinely harder problem than either monolingual case and is worth testing directly rather than assuming.
What to look for when you evaluate it
| Arabic quality measured separately | Not an overall figure. The gap between languages is where deployments disappoint. |
|---|---|
| Formal to informal matching | Whether a colloquial question finds a formally written document. |
| Cross-language matching | Whether Arabic questions reach English material. |
| Domain terminology | Whether your sector's vocabulary is represented sensibly. |
| Robustness to typos | Whether misspelt input still matches, since people type on phones. |
| Behaviour on short input | Whether very short questions match usefully or match everything. |
Where it should not be trusted alone
Similarity is a measure of resemblance, not of correctness. Two passages can be closely related and still not answer the question, and a system that treats high similarity as sufficient will retrieve confidently and answer wrongly.
It is particularly unreliable on very short input. A three-word question resembles a great deal of material weakly, and acting on the nearest match produces answers that feel arbitrary to the person asking.
It should also not be used to conclude that two things are the same for any consequential purpose — that two customer records refer to one person, or that a new complaint duplicates an old one. It can suggest that; a person should decide it.
How this fits the whole
No single capability answers a question on its own. A conversation passes through several in sequence — understanding what was asked, finding the relevant material, composing a reply, checking it before it is sent — and the quality of the result is set by the weakest link rather than the strongest.
That has a practical consequence for evaluation. Vendors tend to demonstrate their strongest component, and a system with excellent retrieval and weak language handling will demo beautifully and disappoint in production, because the demo was conducted in the language and phrasing the system handles best.
It also explains why grounding matters more than raw capability. A more capable model that answers from its own training rather than from your material produces confident, fluent, unsupported answers — which is a worse outcome than a less capable one that retrieves correctly and declines when it cannot. Capability without grounding increases the cost of being wrong rather than reducing it.
The pieces are described separately here because that is how they are evaluated and configured. They are not experienced separately: a customer asking a question encounters one system, and judges it as one.
Where this sits in a conversation
- Something arrivesA typed message, a spoken sentence, an uploaded document. The first job is establishing what was actually received — which language, which script, whether it is legible, whether it is one question or several.
- It is interpretedWorking out what is being asked rather than what words were used. The same question arrives phrased twenty different ways, in two languages, often mixed, frequently misspelt, sometimes as a fragment.
- Material is retrievedThe relevant part of your own approved content is found. This is where answer quality is largely decided: a correct answer assembled from the wrong source is still wrong.
- A reply is composedIn the language of the question, in the configured tone, from the retrieved material and nothing else. Where the material is silent, the reply says so rather than filling the gap.
- It is checkedBefore sending, the reply is examined for claims the retrieved material does not support and for personal data that should not be repeated back. This runs on every reply.
- It is deliveredAs streamed text, as speech, or as a structured response — and the conversation is recorded so that what happened can be reviewed later.
Why this is built rather than bought in
Every capability described here could, in principle, be obtained by calling somebody else's service. That is how most assistants in this market are assembled, and it is why so many of them cannot answer the question that decides regulated procurements: where does the data go.
When a component is a call to an external provider, every question that passes through it leaves the organisation. That may be perfectly acceptable for a retailer answering questions about opening hours. It is usually not acceptable for a bank explaining a transaction, a hospital confirming an appointment, or a public body handling a citizen case — and the constraint is architectural, so it cannot be configured away later.
The second consequence is control over behaviour. A capability obtained as a service behaves the way the provider decides, changes when the provider changes it, and cannot be tuned for the specific way your customers speak or write. A capability that runs inside your own estate can be adjusted for dialect, for your terminology, and for the particular failure modes your material produces.
The third is continuity. External services are deprecated, repriced and withdrawn. A deployment whose core capability depends on somebody else's roadmap inherits that roadmap, including the parts nobody announced in advance.
Measuring this component honestly
| Measured in which language? | Nearly every published benchmark in this field was developed against English. An overall figure that averages across languages hides exactly the gap that matters in this market. |
|---|---|
| Measured on what input? | Clean or realistic. Read speech or spontaneous speech. Typed text or phone-typed text with autocorrect. The difference between these is usually larger than the difference between vendors. |
| Measured by whom? | Self-reported figures with an undescribed method are marketing. Ask for the test set and the procedure. |
| What counts as correct? | A surprising number of published figures count a partially correct answer as correct. Ask what the scoring rule was. |
| How does it fail? | A component that degrades gracefully is worth more than one with a higher headline figure that fails silently and confidently. |
| Does it report uncertainty? | A component that surfaces low confidence lets the system confirm rather than guess. One that hides it produces confident errors downstream. |
What to ask about this specifically
Ask for the Arabic figure
Not the overall figure. Nearly every published benchmark in this field was developed against English, and the gap between the two languages is where deployments disappoint.
Ask how it was measured
Against what test set, by whom, on what kind of input. A number without a described method is marketing rather than measurement.
Ask what it does when uncertain
Whether uncertainty is surfaced or hidden. A component that hides low confidence produces confident errors downstream.
Ask about real conditions
Noisy audio, poor scans, dialect, informal phrasing, mixed languages. Laboratory conditions are not predictive of contact-centre reality.
Ask where processing happens
For any component touching personal data, this is a procurement question rather than a technical curiosity in regulated sectors.
Ask what is retained
What data this component keeps, for how long, and whether it is used to improve anything.
Common questions
It sits underneath retrieval and intent matching. When customers say the assistant "did not understand", the cause is usually here rather than in the reply.
It has to be built for, not translated into. Ask for the Arabic figure separately — an overall number averages away the gap that matters here.
Yes, and it usually needs to, because organisations rarely maintain perfectly parallel content in both languages.
It needs to. People type quickly on phones with autocorrect working against them, and requiring well-formed input fails a large share of genuine questions.
They are the weak case — a three-word question resembles many things weakly. Asking a clarifying question beats acting on a marginal match.
It can suggest it. Concluding it is a decision with consequences and belongs with a person.