Neither lexical search nor learned sparse-vector search is automatically the right choice for multilingual retrieval. BM25 is a strong baseline when queries and documents share a language, analyzer, and vocabulary. Learned sparse models can weight tokens contextually and may add related terms, but language and cross-language support depend on the specific model. For queries and documents in different languages, test translation and multilingual retrieval explicitly; choosing a sparse-vector index alone does not solve the mismatch.
What is the difference between lexical and learned sparse search?
Lexical search: match terms and rank them
BM25 is a lexical ranking method. It scores documents using query-term matches and document statistics, including how often a term occurs and document length. Its behavior depends on how text is analyzed and tokenized before indexing: stemming, segmentation, normalization, and script handling can affect which terms match. OpenSearch documentation describes BM25 in these terms.
When a query and document use compatible language processing and share important terms, this approach gives a clear, dependable baseline. It is also useful for exact names, product codes, and specialist vocabulary when the analyzer preserves those tokens. The BGE-M3 model card notes that BM25 remains competitive, particularly for long-document retrieval.
Learned sparse retrieval: model-weighted token dimensions
A learned sparse model represents text as weighted token dimensions. Unlike ordinary term-frequency scoring, a model estimates the weights; some model families can also activate related vocabulary that the query or document did not literally contain. “Sparse” describes the representation, not a promise of multilingual coverage, semantic accuracy, or cross-language matching.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Model variants differ. NAVER LABS Europe’s SPLADE-v3-Lexical model card labels the model English and describes a 30,522-dimensional representation. BGE-M3 supports sparse retrieval alongside dense and multi-vector modes, and its authors report support for more than 100 languages. These are materially different language claims, so the model’s intended coverage must be checked rather than inferred from the word “sparse.”
Does sparse retrieval work across languages?
Only when the model and retrieval setup support the languages involved. If the query is in one language and the document is in another, ordinary lexical overlap may be limited. A multilingual sparse model trained for cross-lingual retrieval may help, as can query translation or document translation; these are separate strategies whose quality should be tested.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Language-count claims do not establish equal retrieval quality for every language, script, domain, or query style. Common SPLADE variants include English-focused models, while BGE-M3 and OpenSearch multilingual-v1 explicitly target multilingual use. Validate support for the actual language and script combinations in your application.
Keep translation separate from the retriever
Translation changes the retrieval input and can affect results independently of the ranking method. In a French-to-English scientific-document experiment on the Érudit CLIR dataset, BM25 with a French analyzer performed poorly without translation; results varied when translation was introduced. The study by Valentini, Kozlowski, and Larivière (2025) therefore illustrates why a retriever should not be judged apart from its translation setup.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Compare query translation, document translation, and multilingual retrieval under the same corpus and evaluation conditions. Record the translation method and assess translation quality as an experimental variable, rather than attributing every score difference to BM25 or a sparse model.
How do the approaches compare for a multilingual application?
| Decision area | Lexical search with BM25 | Learned sparse retrieval |
|---|---|---|
| Language and script | Needs analyzers and tokenization suited to the languages and scripts being indexed and queried; cross-language term overlap can be limited. | Depends on the particular model’s training and supported languages; a sparse representation alone does not establish multilingual capability. |
| Exact terms and identifiers | Direct term matching is useful when analysis preserves names, codes, and specialist terms. | Model weighting or vocabulary expansion may help related wording, but exact-match behavior should be tested separately. |
| Vocabulary expansion | Ranks matching analyzed terms; it does not itself infer related vocabulary. | Some model families can assign weight to related vocabulary beyond literal input terms. |
| Configuration | Analyzer and tokenizer choices shape matching and need language-appropriate configuration. | Requires compatible model representations for indexing and querying; inference and indexing choices affect reproducibility. |
| Operational requirements | The cited BM25 descriptions establish term-statistic ranking, but do not provide a general comparative index-size or latency figure. | Query inference may be required, or token weights can be precomputed. Elasticsearch sparse-vector query documentation requires query inference to use the same inference model as the indexed tokens. |
| Long documents | The BGE-M3 model card says BM25 remains competitive, especially for long-document retrieval. | Performance depends on the model and task; no general long-document advantage is established by the cited evidence. |
There is no universal winner across these dimensions. Choose based on the languages, content, query patterns, deployment constraints, and retrieval depth that matter to your own system.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
What do published benchmark results show?
The reported figures below come from different datasets and setups. They are useful examples of conditional results, not a shared leaderboard or a prediction of performance on another corpus.
| System and source | Reported result | What the result applies to |
|---|---|---|
| OpenSearch multilingual-v1; OpenSearch Project blog (year not stated in the opened text) | 0.629 average nDCG@10; 0.305 for BM25; 0.626 for multilingual-v1 pruned at ratio 0.1 | Vendor-reported MIRACL results across the listed language tasks. They do not guarantee the same difference on another corpus. |
| BGE-M3 Sparse; Chen et al. (2024) | 0.539 nDCG@10 | MIRACL development set. The same table reports 0.692 for Dense and 0.705 for Multi-vec, showing that retrieval modes within one model can differ. |
| BGE-M3 Sparse and BM25; Valentini, Kozlowski, and Larivière (2025) | 0.575 and 0.638 nDCG@10, respectively | Érudit CLIR French-to-English scientific-document experiment under the GPT-4 query-translation condition. Other translation methods and metrics produced substantial variation. |
| SPLADE-v3-Lexical; NAVER LABS Europe model card (year not stated) | 40.0 MRR@10 on MS MARCO dev; 49.1 average nDCG@10 on BEIR-13 | English-oriented benchmark results. These scores should not be compared directly with MIRACL or CLIRudit because datasets, metrics, and evaluation setups differ. |
Metric cutoffs should match how retrieval is used. nDCG@10 measures ordering near the top of the results; Recall@k measures how many relevant candidates appear by the depth passed downstream. The CLIRudit paper explains why cutoffs can differ between systems that rerank candidates and systems that do not.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
How should you evaluate the options?
Build a fixed, judged test set that reflects the language mix and retrieval task in production. Include queries and documents for each important language and script, plus difficult cases such as rare names, identifiers, and specialist terminology. Those cases can expose weaknesses hidden by averages.
- Configure a lexical baseline. Select analyzers and tokenization appropriate to each language and script. Check that important names, codes, and domain terms survive analysis as intended.
- Select model candidates by actual coverage. Confirm the model variant’s supported language and script use cases. For learned sparse retrieval, use compatible model versions and representations for document indexing and query processing.
- Separate retrieval from translation. Compare query translation, document translation, and multilingual retrieval as distinct configurations. Hold other conditions fixed so translation effects are not mistaken for retrieval effects.
- Freeze the evaluation conditions. Use the same corpus snapshot and judged queries. Record analyzer and tokenizer settings, model checkpoint, translation method, pruning or sparsity controls, and candidate depth.
- Measure both ranking and candidate coverage. Use nDCG@10 for quality near the top when that is the user-facing concern, and Recall@k at the candidate depth the downstream system actually consumes.
- Test hybrid retrieval as a candidate, not an assumption. Combining lexical matches with learned sparse results may cover different failure cases. Keep it only if evaluation shows an improvement for the relevant queries and retrieval depth.
Which approach should you start with?
Start with BM25 when query and document language align, the text can be analyzed appropriately, and exact terminology matters. Add a learned sparse candidate when contextual weighting or vocabulary expansion could address observed misses, and choose a model with documented coverage for the languages and scripts involved. For cross-language retrieval, evaluate translation and multilingual models explicitly. BGE-M3 and OpenSearch multilingual-v1 are candidates to test, not substitutes for local evaluation; BGE-M3’s authors also say generalization to varied real-world datasets needs further investigation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
