AI-102 Implement natural language processing solutions Practice Test 7 — 15 Questions

Question 1

You are using Azure AI Language's text analytics for health to extract medical entities from clinical notes. The notes contain abbreviations and domain-specific terminology. After initial testing, you find that the model misses some entities like 'MI' (myocardial infarction). You need to improve entity recognition without retraining the model. What should you do?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Preprocess the text to expand abbreviations like 'MI' to 'myocardial infarction' before sending to the API.. Preprocessing text to expand abbreviations is an effective way to improve entity recognition without retraining. The model is more likely to recognize full terms than abbreviations. Other options either don't address missed entities or are not feasible.

Answer

Use the model's confidence scores to filter out low-confidence entities and focus on high-confidence ones.

Answer

Submit a support ticket to Azure to add 'MI' to the model's vocabulary.

Answer

Add a custom dictionary of abbreviations and map them to standard terms using the entity linking feature.

Question 2

You are building a solution that uses Azure AI Language to perform sentiment analysis on customer feedback. The feedback is in English and contains multiple sentences per document. You need to obtain both document-level and sentence-level sentiment. Which two actions should you perform? (Choose two.)

Accepted Answer

Set the 'opinionMining' parameter to true in the request body.. To get both document-level and sentence-level sentiment, you should use the analyze-text endpoint with kind set to SentimentAnalysis. Enabling opinion mining further enriches the response with aspect-based sentiment at the sentence level. The deprecated Text Analytics endpoint and manual sentence splitting are not recommended, and there is no portal mode for multi-sentence handling.

Answer

Split each document into individual sentences and send each sentence as a separate document.

Answer

Configure the resource to use the 'multi-sentence' mode in the Azure portal.

Answer

Use the Text Analytics v3.1 endpoint /sentiment with the 'sentenceScores' parameter set to true.

Question 3

You are implementing a conversational language understanding (CLU) project in Azure AI Language for an airline booking bot. The bot must map utterances such as "I want to fly from Seattle to Tokyo next Friday" to an intent named BookFlight and populate a destination entity. Your evaluation run shows high overall precision but the destination entity is frequently missed on utterances that include a connecting city. You need to raise entity recall without degrading intent accuracy. (Choose two.)

Accepted Answer

Review the evaluation results and add labeled examples for utterances where the destination entity was predicted incorrectly.. Entity recall improves when the training data actually contains the failing pattern with correct annotations, and when evaluation feedback drives targeted labeling. Adding multi-city utterances with every city labeled as destination teaches the model the connecting-city context, while reviewing mispredictions and labeling them closes the remaining gaps. Options that only add hints, prebuilt components, or change data splits do not supply the missing labeled signal.

Answer

Add a phrase list that contains the names of all airports served by the airline.

Answer

Enable the prebuilt geography entity component and attach it as a required component of the destination entity.

Answer

Increase the training split percentage in the project settings and retrain the model from scratch.

Question 4

A company is building a customer support bot that must understand user intents and extract entities from utterances. They have created a Conversational Language Understanding (CLU) project in Azure AI Language. They need to train and deploy the model, then query it from a client application. Which endpoint should the client application call to get intent and entity predictions?

Accepted Answer

POST /language/:analyze-conversations. The correct endpoint for querying a Conversational Language Understanding model is /language/:analyze-conversations. This endpoint accepts a request body that includes the kind parameter set to "Conversation" and the query text, and it returns the predicted intent and entities. The analyze-text endpoint is for other text analytics tasks and does not support CLU.

Answer

POST /language/:analyze-text

Answer

POST /language/:analyze-text with kind set to "Conversation"

Answer

POST /language/:analyze-conversations with kind set to "Conversation"

Question 5

You are building an ASP.NET Core web API that receives support tickets. You must detect the language of each ticket and, when the detected language is not English, route the ticket to a separate queue. You want a single REST call that returns the detected language name and its ISO 639-1 code, with a confidence score, without deploying any model. Which Azure AI Language REST endpoint should you call?

Accepted Answer

POST https://<resource>.cognitiveservices.azure.com/language/:analyze-text?api-version=2023-04-01 with kind set to LanguageDetection. The unified Azure AI Language analyze-text endpoint accepts a kind parameter that selects the analysis task. Setting kind to LanguageDetection returns the language name, ISO 639-1 code, and confidence score synchronously, which matches the need to branch tickets by language. The other endpoints serve conversational understanding, question answering, or a legacy text analytics path and do not provide standalone language identification.

Answer

POST https://<resource>.cognitiveservices.azure.com/text/analytics/v3.1/languages

Answer

POST https://<resource>.cognitiveservices.azure.com/language/:analyze-conversations?api-version=2023-04-01

Answer

POST https://<resource>.cognitiveservices.azure.com/language/:query-knowledgebases?api-version=2023-04-01

Question 6

You are building a conversational language understanding (CLU) project in Azure AI Language. You need to train a model that can identify intents and extract entities from user utterances. The project must support English and Spanish. What should you do to ensure the model can handle both languages?

Accepted Answer

Add utterances in both English and Spanish to the same intent and entity definitions.. CLU supports multilingual projects by allowing utterances in multiple languages within the same intent and entity definitions. This enables the model to learn from both languages without separate projects or translation. Creating separate projects or translating all data adds unnecessary complexity and may reduce accuracy.

Answer

Create separate CLU projects for each language and combine the results at runtime.

Answer

Use the Azure AI Translator to translate all utterances to English before training.

Answer

Set the project language to 'multilingual' in the Azure portal.

Question 7

You are designing an Azure AI Language solution that must detect personally identifiable information (PII) in customer chat transcripts and redact it before storage. The transcripts are in English and contain names, phone numbers, email addresses, and physical addresses. You need to configure the PII detection feature in Azure AI Language. (Choose two.)

Accepted Answer

Specify the language parameter as en so the model applies English PII categories and confidence scoring. The solution must identify PII in English transcripts and redact it before storage. Declaring the en language ensures the correct model and categories are applied, while the redaction parameter replaces detected entities with a masking character in the returned text, satisfying the redaction requirement in a single call. The phi domain, opinion mining, and string index type are valid parameters but are not required for this general customer chat scenario.

Answer

Enable the opinion mining parameter so the service distinguishes subjective statements from factual ones in the transcript

Answer

Set the stringIndexType parameter to TextElements_v8 to control how entity offsets are calculated in the response

Answer

Call the analyze-text endpoint with the PiiEntityRecognition task and set the domain parameter to phi for health-related content

Question 8

You are implementing a custom named entity recognition (NER) solution using Azure AI Language. You have labeled data and need to train and deploy a model. You want to ensure the model can recognize entities in production. Which two actions should you perform? (Choose two.)

Accepted Answer

Train the model using the Language service's training API or Language Studio, then review the model evaluation metrics before deployment.. To implement custom NER, you must create a CustomEntityRecognition project and import labeled data in the supported format, then train the model and evaluate its performance. Deployment uses a named deployment rather than slots, there is no multi-lingual toggle, and prebuilt models cannot detect custom entities. Training and proper project setup are the essential actions.

Answer

Enable the 'multi-lingual' option in the project settings to support all languages automatically.

Answer

Use the prebuilt entity recognition endpoint to detect your custom entities without training.

Answer

Deploy the model to a slot named 'staging' and use the staging endpoint for production traffic.

Question 9

You are developing a conversational language understanding (CLU) project in Azure AI Language. The project must recognize user intents and extract entities from utterances in both English and French. You have created a project and added utterances in both languages. You need to ensure that the model can be trained and evaluated for both languages. What should you do?

Accepted Answer

Create separate CLU projects for English and French, each with the appropriate language setting, and train them independently.. Azure AI Language's conversational language understanding requires a single language per project. To support both English and French, you must create separate projects for each language. Each project is trained with utterances in its respective language. This ensures the model learns language-specific patterns and entities correctly. Attempting to mix languages in one project is not supported and will lead to poor accuracy.

Answer

Add utterances in both languages to a single project and set the language to "multilingual" in the project settings.

Answer

Use the multilingual option when creating the project and select both English and French.

Answer

Set the project's primary language to English and add French as a secondary language in the project settings.

Question 10

You are building a solution that uses Azure AI Language's personally identifiable information (PII) detection to redact sensitive data from customer feedback. You need to ensure that the redaction replaces detected PII with a category label, such as [Person] or [PhoneNumber], instead of asterisks. Which parameter should you configure in the API request?

Accepted Answer

Set the 'redactionPolicy' to 'MaskWithEntityType'.. To replace detected PII with category labels like [Person] or [PhoneNumber], you must set the redactionPolicy to MaskWithEntityType. This policy uses the entity type as the replacement text. Other parameters like piiCategories filter detection, and domain adjusts the model, but they do not control the redaction format.

Answer

Set the 'redactionPolicy' to 'MaskWithCharacter' and specify the character.

Answer

Set the 'piiCategories' parameter to include only the entities to redact.

Answer

Set the 'domain' parameter to 'phi' to use healthcare-specific redaction labels.

Question 11

You are using Azure AI Language's custom question answering to build a bot that answers questions from a knowledge base of product manuals. The knowledge base contains multiple languages, and you need to ensure that users can ask questions in their own language and receive answers in that language. You have already created the project and added sources. What should you do to enable multilingual support?

Accepted Answer

Enable the multilingual option in the project settings and specify the default language.. Custom question answering includes a multilingual capability that can be enabled in project settings. When enabled, the service automatically detects the language of each question and returns answers from the knowledge base in that language, provided the sources include content in that language. This avoids the need for separate projects or external translation.

Answer

In the project settings, set the language to 'multilingual' and add language-specific sources.

Answer

Create a separate project for each language and use a dispatcher to route questions.

Answer

Add a translation layer using Azure AI Translator before and after calling the question answering API.

Question 12

A healthcare company stores clinical notes in Azure Blob Storage. You must identify protected health information such as patient names, ages, and medical record numbers in the notes. You need a solution that returns the recognized entities with their categories and offsets, and that lets you exclude specific categories from the response. Which Azure AI Language capability should you implement?

Accepted Answer

Named entity recognition (NER) with the prebuilt healthcare domain model, called via the analyze-text endpoint using kind set to EntityRecognition and domain set to phi. The prebuilt healthcare domain of named entity recognition is purpose-built for clinical text and returns categorized entities with offsets and confidence scores, including PHI categories. Selecting it through the analyze-text endpoint with the appropriate domain value gives the compliance team the structured output it needs without the cost and labeling effort of a custom model, and the response can be constrained to the categories of interest.

Answer

Key phrase extraction on each clinical note to surface the most relevant terms

Answer

Personally identifiable information (PII) detection with the default domain on the analyze-text endpoint

Answer

Custom named entity recognition trained on a subset of labeled clinical notes

Question 13

A company wants a bot that answers employee questions from a large set of HR policy documents. The documents change frequently and the team wants answers composed from the most relevant passages rather than from a fixed list of question-and-answer pairs. Which Azure AI Language capability should you use?

Accepted Answer

Question answering with a knowledge source that indexes the HR documents and generates answers from the top passages.. Question answering with a document-backed knowledge source retrieves the strongest passages and can compose an answer from them, which suits a corpus that changes often. Because the knowledge source is indexed from the documents themselves, updates flow through reindexing rather than through manual authoring of question-and-answer pairs, aligning with how the HR team maintains its policies.

Answer

Custom named entity recognition trained on HR policy terminology.

Answer

Conversational language understanding with an intent for each HR topic.

Answer

Custom question answering with a project built from the HR documents.

Question 14

You are designing a solution that uses Azure AI Language's question answering (custom question answering) to build a FAQ bot. The bot must answer questions from a knowledge base that contains both structured and unstructured documents. You need to ensure that the bot can provide answers with high confidence and also support follow-up prompts. Which two features should you configure? (Choose two.)

Accepted Answer

Add multiple turns to question-answer pairs to create follow-up prompts.. Active learning and multiple turns are the two features that directly address the requirements. Active learning improves the knowledge base over time by suggesting alternative questions, which increases confidence. Multiple turns enable follow-up prompts, allowing the bot to guide users through a conversation. Together, they enhance the bot's ability to answer questions accurately and interactively.

Answer

Set the confidence threshold to 0.9 to ensure only high-confidence answers are returned.

Answer

Import the documents using the file upload feature and enable automatic extraction of question-answer pairs.

Answer

Configure the bot to use the precise answering feature for all questions.

Question 15

You are using Azure AI Language's sentiment analysis to analyze customer feedback. You need to determine the sentiment at both the document level and the sentence level. Which parameter should you set in the API request to get sentence-level sentiment?

Accepted Answer

Set 'showSentenceSentiment' to true.. The sentiment analysis API in Azure AI Language provides sentence-level sentiment when the 'showSentenceSentiment' parameter is set to true. This returns an array of sentences with their individual sentiment scores. The other options are either incorrect parameter names or unrelated features like opinion mining.

Answer

Set 'opinionMining' to true.

Answer

Set 'sentenceSentiment' to true.

Answer

Set 'includeSentenceSentiment' to true.